Stock Buybacks: When They Lift EPS and When They Only Offset Dilution
Booking Holdings reported net income 8.1% lower in 2025 than in 2024. Its diluted earnings per share fell only 4.1%. The four points between those two numbers are not an accounting trick. They are shares that no longer exist, retired with cash the company spent buying its own stock.
That is the good version of a buyback, and it is the version everyone imagines when a company announces a new authorization. The bad version looks identical in a press release. A company spends billions on repurchases, and the share count on the balance sheet sits exactly where it was, because the buying merely absorbed the stock that employees received as pay. Cash left the business, nobody’s ownership stake grew, and the headline still said “returns capital to shareholders.”
My view is simple enough to state before the evidence. A buyback deserves credit only when the diluted share count falls over several years, when the money came from free cash flow rather than new debt, and when the price paid leaves the owner ahead of the alternatives. The announced dollar figure tells you almost nothing about any of those three.
A one-line test using numbers you can find
You do not need a repurchase schedule to see whether buying is working. Take net income growth and diluted EPS growth for the same year. If EPS grows faster, the average share count fell, and the size of the difference is roughly the percentage of shares retired. If EPS grows slower, the count rose.
I ran that on twelve large companies using annual figures from a market database, then divided net income by diluted EPS to back out an implied share count and compared five years apart. The implied count is approximate, because net income and diluted EPS are not always built on exactly the same basis, but the direction and rough size hold up.
| Company | Net income growth | EPS growth | Gap (points) | Implied share count, five years |
|---|---|---|---|---|
| O’Reilly Automotive | +6.3% | +9.6% | +3.2 | -23.5% |
| Booking Holdings | -8.1% | -4.1% | +4.0 | -20.3% |
| Apple | +19.5% | +22.7% | +3.2 | -14.2% |
| Linde | +4.8% | +7.3% | +2.5 | -13.4% |
| Alphabet | +32.0% | +34.5% | +2.4 | -11.0% |
| Meta Platforms | -3.0% | -1.6% | +1.5 | -10.9% |
| Mondelez | -46.7% | -44.7% | +1.9 | -9.7% |
| Monster Beverage | +26.3% | +30.2% | +3.9 | -8.0% |
| Thermo Fisher | +6.0% | +7.3% | +1.3 | -5.2% |
| Costco | +9.9% | +10.0% | +0.0 | -1.2% |
| Southern Company | -2.1% | -1.8% | +0.3 | +0.5% |
| Newmont | +112.0% | +118.8% | +6.9 | +41.1% |
Where the shrinking count is real
O’Reilly is the cleanest example in the set. The implied count fell from roughly 1,117 million shares in 2020 to roughly 854 million in 2025, a 23.5% cut, and diluted EPS rose from $1.57 to $2.97 while net income rose from $1.75 billion to $2.54 billion. Profit grew about 45% over those five years. Per-share profit grew about 89%. The repurchases account for the difference, and an owner who simply held the stock and did nothing collected all of it.
Booking is the more interesting story to me, because it shows the mechanism working while the business stumbled. Profit went backward in 2025 and the per-share number cushioned the fall. I would not read that as strength. A cushion is not a recovery, and if profit keeps sliding, shrinking the denominator only buys time.
Apple sits between them. Its count fell about 14% in five years, which meant EPS grew roughly 3 points faster than profit last year. I wrote earlier about why the buyback matters more than unit growth for Apple, and the arithmetic supports that: at Apple’s scale a 3% annual reduction in shares is worth more to EPS growth than most consumer hardware categories add in a good year.
Linde, Alphabet and Thermo Fisher fit the same pattern at different speeds. What they share is a business that throws off more cash than it can reinvest at an attractive return. When a company like that sends the surplus back through repurchases, the count falls, and the loop is healthy provided the operating profit keeps growing underneath.
Where the money leaves and the count stays
Southern Company is my favorite small example, because it is unglamorous. Its implied count moved from about 1,059 million shares to about 1,064 million over five years, up half a percent. A regulated utility needs capital for its investment plan, and it raises some of it by issuing stock. Nobody should read that as a failure. The dividend is the return story there.
Costco is the version that asks a harder question. Its gap last year was 0.0 points and its implied count fell only 1.2% in five years. Costco does repurchase stock, so the money left the building. But the shares it issues to employees and executives absorb most of it. The owner’s slice of the company barely changed. If you paid 45 times earnings for the stock as I write this, and then saw the buyback described as a capital return, you were not being told what happened.
Newmont breaks the tidy pattern in an instructive way. Its one-year gap was the largest in my table, 6.9 points, which would suggest a heavy buyback. The five-year count says the opposite: up about 41%. I attribute most of that to shares issued in its stock-funded acquisition of Newcrest, not to stock pay. The point is that a single-year gap can flatter a company whose count exploded a few years earlier. Always look at more than one period.
Monster Beverage is the one that corrects a common assumption, that a company with a small buyback must have a rising share count. The database says the implied count is down about 8% over five years and its EPS grew about 4 points faster than net income last year. Small buyback as a share of profit, perhaps, but a visible effect.
Perceptions of a buyback are often wrong, and the share count settles the argument.
What the price paid does to the arithmetic
Same dollars, different result. Suppose a company earns $1 billion a year and has 100 million shares, so EPS is $10. It spends $1 billion on repurchases. If the stock trades at 40 times earnings, or $400, that buys 2.5 million shares, a 2.5% reduction. If it trades at 15 times, or $150, the same money buys 6.7 million shares, a 6.7% reduction. The earnings yield on the buyback is the inverse of the multiple, and it is the return the owner earns on that cash.
That is why I get uneasy when a richly priced company announces a giant program. At 40 times earnings, retiring stock earns you a 2.5% yield on the cash, and the interest rate on a money market fund may beat it. The buyback has value. It is priced like a mediocre investment and presented like a generous one.
There is also a quiet tax angle in the United States. A 1% federal excise tax on net repurchases began with buybacks in 2023, applied to the net amount after subtracting shares issued. It is small against a big program, yet it does mean the offset math (buying to cover employee grants) is now slightly more expensive than it used to be.
Debt-funded buying is a different animal
Free cash flow is the honest source of buyback money. When repurchases exceed free cash flow for several years running, the gap gets filled with debt, and the shrinking share count arrives together with a rising interest bill. Booking is a case where I would want to check that. A company that borrows to retire shares is making a borrowed bet that its own stock is cheap, and the return on that bet depends entirely on the price paid. The bank posts I wrote on capital return make the same point in a different industry: a distribution that rests on regulatory capital is not the same as one that rests on surplus cash.
I would also not lean on authorizations. An authorization is permission to buy, not an order. Companies announce large numbers, buy a fraction when the stock is high, and hold the rest in reserve. The cash-flow statement shows what actually went out, and the share count shows what that bought.
Why a one-year gap can mislead
The gap has limits I should state plainly. A one-time gain or loss moves net income and EPS together, but not always by identical proportions. Tax changes can distort the year. A swing in operating profit can swamp the share effect. Mondelez fell 46.7% in net income last year, and that gap of 1.9 points is nearly meaningless next to the size of the profit drop. So I use the number as a screen. It tells me where to look. It hands down no verdict.
The main risk to my framework is a company whose gap is persistently near zero but whose stock keeps climbing because profit compounds at 20% a year. That would prove me wrong, because the buyback would not matter to the return and I would be spending attention in the wrong place. Costco is not far from that case.
The one line I would open first
If I had five minutes on a stock that a friend swore was a great buyback story, I would open the income statement and read diluted weighted-average shares for the last five years. If the number has dropped by 3% or more a year on average, the program is real. If it has moved by less than 1% a year, I would treat the announcements as marketing and look for a different reason to own the stock. That threshold is my rule of thumb, not a law, and a company with a very high stock price and modest compensation can clear it with a smaller program.
Analysis and opinion only, not investment advice. Figures come from annual company data in a market database and from filings on SEC EDGAR for O’Reilly Automotive and Newmont; implied share counts are my own approximation, and everything was checked on September 22, 2026.