← Quantery Blog

Share count and dilution: why per-share numbers drift

August 31, 2026 · 8 min readfundamentalseducation

Every per-share number is a fraction, and most of the attention goes to the top of it. Earnings per share, book value per share, free cash flow per share: a business result on top, a share count underneath. The count underneath moves on its own schedule. Companies print shares to pay employees, to buy other companies and to raise cash, then buy shares back to shrink the count again. So a per-share number can travel a long way while the business behind it does nothing at all.

Two counts show up in the filings and they answer different questions. Basic shares are the weighted average number of shares actually outstanding across the reporting period. Diluted shares add the shares that would exist if the options, restricted stock units and convertible securities already granted turned into stock today. Diluted is the bigger and more conservative figure, and it's the one the bundled theses read, because a claim on the business that hasn't been exercised yet is still a claim on the business.

This series builds up the vocabulary our theses use, one value at a time. An earlier piece walked a balance sheet end to end, where equity sits as the residual claim of the shareholders. Divide that residual by the share count and you have book value per share, which is where the denominator starts to matter.

Where the share count lives in a filing

Three places, and they disagree with each other by design.

The income statement carries the weighted average counts and the per-share results. Regulation S-X lists the captions a company has to present there, and caption 25 is "Earnings per share data". Basic and diluted sit side by side, each with its own share count under it.

The cover page of the 10-K or 10-Q gives a point-in-time count: shares outstanding as of a date a few weeks before the filing goes out. Nothing is averaged.

The cash flow statement shows the money moving, down in the financing section: proceeds from issuing stock on one line, cash spent repurchasing it on another. The statement of stockholders' equity rolls the count forward from the start of the period to the end, one cause at a time. That's the statement to open when you want to know why the count changed.

The weighted-average versus point-in-time split is where per-share comparisons go wrong. A company that doubles its share count on the last day of a quarter barely moves the weighted average for that quarter, while its cover-page count doubles. Quarterly EPS comes out looking almost untouched. Book value per share, computed off the balance sheet date, has halved. Both numbers are correct. They're answering different questions, and reading one when you wanted the other will hand you a per-share trend that never happened.

Dilution's cost depends on the price the shares fetched

AMC in 2021 is the clearest recent case in the public record, and the company's own announcements carry the numbers.

Between 14 December 2020 and 25 January 2021, AMC raised or secured commitments for $917 million. The equity part of that: $506 million from the issuance of 164.7 million new common shares, plus another 22 million shares handed over to convert $100 million of second-lien debt into stock. Call it 186.7 million shares created in roughly six weeks.

For scale, AMC later reported 501,780,240 shares outstanding and entitled to vote as of 2 June 2021. Those six weeks of issuing account for about 37 percent of that count.

Now the same company, five months on. On 3 June 2021 AMC completed an offering of 11.55 million shares that brought in about $587.4 million, roughly $50.85 a share. Put the January arithmetic beside it: $506 million spread across 164.7 million shares is about $3.07 a share. Sixteen times the cash per unit of dilution, same company, five months apart.

Dilution is measured in what the shares bought, not in how many were printed. An owner diluted in January gave up a slice of the company for about three dollars of cash coming in the door. An owner diluted in June gave up the same slice for fifty. On a share-count basis those look like the same event, and they are nothing like the same transaction.

None of which says the January raises were a mistake. A company running out of cash doesn't get to hold out for a better price, and the alternative to selling stock at three dollars may well have been a bankruptcy filing. That's the limit of the field: it records that ownership moved and stays silent on whether the trade was worth making.

Buybacks are the same trade run backwards

A repurchase shrinks the denominator. Net income holds still, the count falls, EPS rises. The business earned no additional dollar; the claim attached to each share got bigger.

The price dependence runs the same way, pointed the other direction. A company buying its own stock is making a purchase like any other, and it's a good purchase only when the stock is worth more than what it costs. The SEC makes issuers put that reasoning in writing: Item 703 of Regulation S-K requires disclosure of "the objectives or rationales for each repurchase plan or program and the process or criteria used to determine the amount of repurchases". Read that paragraph before you treat a buyback as good news.

Gross buyback dollars are the wrong yardstick anyway, because much of the repurchasing in any given year exists to soak up the shares handed to employees. Net issuance settles it in one subtraction: share count now against share count a year ago. If a company repurchased heavily all year and the count came out flat, the buyback bought back the compensation.

Which is also where free cash flow has a blind spot. Stock-based compensation gets added back to operating cash flow as a non-cash expense, so FCF never charges the business for paying people in stock. The bill arrives in the denominator instead. That's why FCF per share across several years tells you something the FCF line by itself won't.

How the bundled theses read the share count

Two of the five shipped templates watch dilution, for different reasons.

The Piotroski F-Score template treats issuance as a financing signal. The logic in the original paper: a company in decent shape funds itself out of operations, so raising equity while the fundamentals are weak is the company telling you something about its own runway. One of the template's eight signals is the absence of new issuance.

shares_now:   newest(shares_diluted)
shares_prior: lag(shares_diluted, 4)
f_shares:     shares_now <= shares_prior * (1 + $dilution_tolerance)

lag(shares_diluted, 4) reaches back four quarterly filings, so the comparison lands on the same fiscal quarter a year earlier (why the same quarter matters here too). The dilution_tolerance parameter ships at 0.02. Two percent of year-over-year share growth still passes, because a normal company hands out restricted stock units every quarter, and a rule that fired on every RSU grant would fire on almost everybody. Above that line the signal goes false and the name drops a point out of eight.

Graham Net-Net reads the same field for a different worry. Net-nets are tiny companies priced below their current assets minus everything they owe, and the usual way that floor disappears is a financing round:

dilution:
  rules:
    - { when: "is_null(shares_now) or is_null(shares_prior)", score: 0, flag: no_share_history }
    - { when: "shares_now <= shares_prior * (1 + $dilution_max)", score: 2 }
    - { when: "shares_now <= shares_prior * 1.10", score: 1 }
    - { else: 0 }

The template's own comment puts it bluntly: microcaps fund burn by printing shares, and the dilution transfers the asset value you thought you bought to whoever buys the next offering. Note the first rule. A missing share history scores zero and raises a flag, so a company that can't be measured never earns credit for being unmeasurable.

Writing your own dilution rule

Once you can read the two counts, the rule is short. The shape below is illustrative and the bundled templates are the reference for exact fields:

# "Is the pie being cut into more slices?" - illustrative
params:
  drip_tolerance:  0.02      # RSU drift that doesn't count as issuance
  buyback_credit: -0.02      # count shrinking 2%+ over the year

features:
  shares_now:   newest(shares_diluted)
  shares_prior: lag(shares_diluted, 4)
  share_growth: if(shares_prior > 0, shares_now / shares_prior - 1, null)
  equity:       newest(total_equity)
  bvps:         if(shares_now > 0, equity / shares_now, null)

criteria:
  dilution:
    rules:
      - { when: "is_null(share_growth)", score: 0, flag: no_share_history }
      - { when: "share_growth <= $buyback_credit", score: 2 }
      - { when: "share_growth <= $drip_tolerance", score: 1 }
      - { else: 0 }

gate:
  - share_growth <= 0.10     # no runaway issuance

Notice what the parameters don't settle. Whether a shrinking count deserves the top score is a live question: one reading calls it capital discipline, another calls it a company with no better use for its cash than buying itself. Those are two theses about one field, and they'll surface different companies. Write both and test both, which beats arguing about it. The DSL reference has the full field vocabulary.

What the share count can't tell you

The count is arithmetic. The interpretation isn't.

A biotech printing shares to fund a trial and a failing retailer printing shares to make payroll leave identical marks in the denominator. So does a company issuing stock to buy a competitor. So does a three-for-one stock split, which changes everything about the number and nothing about the business, and which shows up as a jump when your data is the filings as they were filed at the time. A rule that reads share growth catches all of these and grades none of them.

Per-share drift is a question with a cheap answer. Open the statement of stockholders' equity and find out which cause moved the count: options exercised, an acquisition paid in stock, an at-the-market program, a buyback. The filing names the cause. The count only reports that something happened.

Build the rule, see which companies it catches, then go read what they did with the money.

Next in the series: revenue, net income and margins on the income statement.

Want to try this on your own rules? Quantery is free for 14 days: the full app, no card required.

← All articles