How to Strip One-Time Gains Out of Earnings Before Valuing

What would you make of a large, steady healthcare company whose earnings per share more than doubled in a single year, and then fell by half the year after? Most readers would assume something dramatic happened to the business. A blockbuster product, perhaps, followed by a patent cliff. Or a recession that arrived on schedule.

Nothing like that happened to Abbott Laboratories. According to Abbott's 10-K for fiscal 2024, diluted earnings per share were $7.64. The year before they had been $3.26, and in fiscal 2025 they came in at $3.72. Pretax income across those three years moved in a narrow, unremarkable band. The swing in the bottom line came almost entirely from one line on the income statement: income taxes.

That's the trap I want to deal with here. A P/E ratio divides price by earnings, and it has no way of knowing which earnings will come back next year. Feed it a year inflated by a gain that won't recur and it hands you a multiple that looks cheap, with nothing cheap about the business. I want to walk through Abbott as a case, then turn it into a method you can run on any company before applying a multiple.

Abbott's Fiscal 2024: The Earnings Line That Doubled on Its Own

Start with the pretax numbers, because they tell you what the business itself earned. From Abbott's 10-K filings, pretax income was $6,664 million in fiscal 2023, $7,013 million in fiscal 2024 and $8,466 million in fiscal 2025. That's growth of about 5% and then about 21%. Operating income followed a similar path: $6,478 million, $6,825 million and $8,053 million. A good business having a decent run. Nothing more.

Now look at the tax line. In fiscal 2023, Abbott recorded income tax expense of $941 million. In fiscal 2025 it recorded $1,942 million. In fiscal 2024, between them, it recorded a tax benefit of $6,389 million. That's a negative number in the expense line, and it was large enough that reported net income of $13,402 million came out almost twice as large as the $7,013 million of pretax income it was built on.

Divide the tax line by pretax income and the oddity is plain:

• Fiscal 2021: $1,140 million ÷ $8,211 million = 13.9%
• Fiscal 2022: $1,373 million ÷ $8,306 million = 16.5%
• Fiscal 2023: $941 million ÷ $6,664 million = 14.1%
• Fiscal 2024: −$6,389 million ÷ $7,013 million = −91.1%
• Fiscal 2025: $1,942 million ÷ $8,466 million = 22.9%

A company doesn't run at a negative 91% tax rate as a matter of course. Something discrete happened, and the 10-K's income tax footnote is where Abbott explains what it was. For valuation purposes I'm going to treat that classification as an input rather than audit it. Whether the benefit was booked correctly is an accounting question, and a separate one. The question here is narrower: should a valuation multiple be applied to it? It shouldn't. Not if the multiple is meant to capitalize earnings the business can repeat.

And notice what the reported numbers say about the following year. Diluted EPS fell from $7.64 to $3.72, a drop of about 51%. A screen sorting by earnings growth would flag fiscal 2025 as a collapse. The pretax line says the opposite: the business earned about 21% more before tax than the year before.

Rebuilding a Clean Earnings Base in Four Steps

The repair is mechanical once you know where to look. I run it in this order, and the order matters: you adjust the pretax line first and tax it second, so that each adjustment carries its own tax effect instead of being double-counted.

Clean earnings = (Reported pretax income − Non-recurring pretax gains + Non-recurring pretax charges) × (1 − Normal tax rate)

1. Find the discrete items. Read the income statement for lines that don't appear every year (gain on sale of a business, litigation settlement, impairment, debt extinguishment), then read the MD&A and the tax footnote for items folded into larger lines. A tax benefit almost never gets its own line; it shows up as an effective rate that makes no sense.
2. Adjust pretax income. Subtract one-time gains and add back one-time charges. Be symmetric: if you strip a gain from the good year, you have to add back the impairment from the bad one, or your clean earnings will drift low over time.
3. Choose a normal tax rate and apply it. The usual candidates are the average effective rate over the surrounding years, the rate in the nearest clean year, or the rate management guides to. Name which one you used.
4. Divide by diluted shares, and compare clean EPS with the reported figure. The ratio between them tells you how much of the headline multiple is real.

Here's Abbott run through those steps. Step 1: the discrete item is the fiscal 2024 tax benefit. Step 2: I'm not adjusting pretax income, because the distortion sits below it. (I'll come back to why that's a simplification.) Step 3: I'll use two normal tax rates and keep both, which is the honest way to handle a choice I can't resolve with certainty.

• Rate A, the three-year average before the benefit: ($1,140 + $1,373 + $941) million ÷ ($8,211 + $8,306 + $6,664) million = $3,454 million ÷ $23,181 million = 14.9%
• Rate B, the next clean year: fiscal 2025's 22.9%
• Clean fiscal 2024 earnings at Rate A: $7,013 million × (1 − 0.149) ≈ $5,968 million
• Clean fiscal 2024 earnings at Rate B: $7,013 million × (1 − 0.229) ≈ $5,404 million
• Step 4, divided by 1,748 million diluted shares: about $3.41 per share at Rate A and $3.09 at Rate B, against the reported $7.64

So reported EPS overstated clean EPS by a factor of roughly 2.2 to 2.5. The gap at Rate A, about $7.4 billion, is simply the difference between the tax Abbott would have paid at its normal rate and the benefit it actually recorded. That's $4.25 a share of earnings with no claim on the next year.

Now apply a multiple, and keep price and value apart while you do it. Price is what the market quotes; value is what the earnings stream is worth. Take a share price of $120, a round number chosen for the arithmetic and not a quote. On reported fiscal 2024 EPS that's a P/E of about 15.7×. On clean EPS it's about 35.2× at Rate A and 38.8× at Rate B. Same price, same business, and the ratio more than doubles depending on which earnings figure sits underneath it. Someone who screened for a sub-16 P/E would have bought a multiple that wasn't there. Our explainer on the P/E ratio and what it doesn't tell you covers this blind spot in general terms; this is what it looks like in a real filing.

The clean series also reads sensibly over time. Clean fiscal 2024 EPS of $3.41 at Rate A sits between $3.26 in fiscal 2023 and $3.72 in fiscal 2025. Growth of about 5% and then 9%. That's the business you'd actually be valuing.

Where the Method Gets Harder Than It Looks

Abbott is a clean case precisely because the distortion was so large and so obviously isolated. Most real cases are messier, and I'd rather name the weak spots than pretend the four steps settle everything.

Choosing the normal tax rate

I'm less sure about this step than I'd like to be. Rate A and Rate B differ by eight percentage points, which moves clean fiscal 2024 EPS by about 10%. The three-year average reflects Abbott's history; the fiscal 2025 rate may better reflect where the company's tax position sits now. I don't think either is wrong. When the choice moves the answer this much, carry both into the valuation as a range rather than splitting the difference and pretending to precision you don't have.

Recurring non-recurring items

Some companies report a restructuring charge, an impairment or a legal settlement nearly every year, each one labeled as unusual. Strip them all out and you've built an earnings base the company has never actually delivered. My rule: if an item labeled one-time shows up in three of the last five years, treat its average as a recurring cost. The label is management's; the decision about what recurs is yours.

Gains you can't see on the face of the statement

Gains on asset sales often sit inside other income or even inside operating income, and a large litigation recovery can be netted against expenses. That's why step 1 reads the MD&A and the footnotes rather than just the income statement. In the Abbott example I didn't adjust pretax income, but a careful analyst would still read the filing for smaller discrete items buried in it. I'd treat my clean figure as the upper bound of the adjustment, not a finished number.

One-time gains versus cyclical peaks

This method handles discrete events. It doesn't handle a business earning above its long-run average because the cycle is favorable; that needs the different treatment in our guide to normalizing earnings for a cyclical business. The test is whether the item would recur under the same conditions. A tax benefit from a one-off event won't. A commodity producer's peak margin might, once the cycle turns again.

The same pattern appears elsewhere, by the way. Tesla's 10-K for fiscal 2023 shows pretax income of $9,973 million and an income tax benefit of $5,001 million, for net income of $14,997 million. In fiscal 2024, pretax income was $8,990 million, down about 10%, but net income fell to $7,091 million, down about 53%. Read only the bottom line and you'd conclude profits halved. They didn't. A one-time item rolled off.

If you'd rather avoid single-year earnings altogether, the approach in earnings power value versus a DCF builds the base from sustainable operating profit, which forces this cleanup by design. Either way, the cleaned number then feeds into the broader process in our guide to calculating the fair value of a stock.

Key Takeaways

• A P/E ratio has no way of telling repeatable earnings from a one-off gain. Abbott's fiscal 2024 tax benefit of $6,389 million pushed diluted EPS to $7.64, while clean EPS at a normal tax rate was roughly $3.09 to $3.41.
• Adjust pretax income first, then apply a normal tax rate, and be symmetric: strip one-time gains and add back one-time charges.
• When the choice of normal tax rate moves the result materially, carry both versions forward as a range. Fair value is a compass, not a GPS coordinate, and a clean earnings base is an estimate too.
• An item labeled one-time that recurs most years is a cost. Treat its average as part of normal earnings.