The last unfinished drill is the simplest to compute and the hardest to compute honestly. In his 1984 letter to Berkshire Hathaway's shareholders, Buffett set out a standard for judging whether a company should retain earnings at all: each dollar retained should produce at least one dollar of market value. Below that, shareholders would have been better off with the cash.
The naive implementation is to compare the change in market capitalisation over a decade with the cumulative earnings retained over the same decade. The naive implementation is wrong, and it is wrong in a way that flatters the wrong companies, because it ignores that a buyback reduces market capitalisation by removing shares while returning value to the holders who stayed.
The adjusted form is one line:
retained earnings = cumulative net income - cumulative dividends
market value created (adj) = (ending market cap + cumulative buybacks)
- (beginning market cap + equity issued)
one-dollar ratio = market value created (adj) / retained earningsAdvertisement
Worked example: IBM and a synthetic manufacturer
Run it on a decade that failed. IBM between 2010 and 2019 earned roughly $125 billion, paid roughly $50 billion of dividends, and therefore retained roughly $75 billion. It repurchased roughly $88 billion of its own stock and issued roughly $3 billion through share-based compensation. Its market capitalisation went from roughly $190 billion to roughly $120 billion. (Figures rounded to teaching scale from the published filings and checked in September 2026; check against the 10-K record before quoting any of them.)
The raw reading is a ratio of −0.93, which is meaningless because it double-counts the buyback. The adjusted reading is (120 + 88) − (190 + 3) = $15 billion of value created against $75 billion retained, a ratio of 0.20. Each dollar retained produced twenty cents of market value, and $60 billion of shareholder capital was consumed in the process. Flex every input by 10% in the company's favour and the ratio is still 0.96; the verdict does not depend on the rounding.
Now a decade that passed. A synthetic manufacturer earned $9.20 billion, paid $2.30 billion of dividends, retained $6.90 billion, repurchased $3.40 billion, issued $0.25 billion, and grew its market capitalisation from $6.10 billion to $22.40 billion. The adjusted value created is $19.45 billion, a ratio of 2.82. That is an unambiguous pass, and it is where most people stop.
Two caveats stop them being right.
What are the caveats of the one-dollar test?
The first is the multiple did some of the work. That company's earnings went from $0.50 billion to $1.40 billion while its multiple went from 12.2 times to 16.0 times. Hold the multiple at its starting level and the ending market capitalisation would have been $17.08 billion, the adjusted value created $14.13 billion, and the ratio 2.05. So 0.77 of the 2.82, or 27.35% of the reading, came from investors changing their minds about what the earnings were worth, and not from anything management allocated. Re-rating flatters a good allocator and damns a bad one, and both are the market's doing.
The second is the window. End the same test one year earlier, at a market capitalisation of $14.60 billion after a drawdown, with $3.10 billion of cumulative buybacks and $6.10 billion retained, and the ratio is 1.86 instead of 2.82. One year of window moved the verdict by 0.96 of a point, or 34% of the reading. The test needs a long window precisely because the market's own error is a large part of the numerator over any short one, and a decade is close to the minimum that makes it worth computing.
Used with both caveats, the one-dollar test is the single most compressed statement of a decade of capital allocation available. Used without them, it is a way of grading management on the market's mood.