What the F-Score measures, and what it does not
The F-Score measures whether a company's fundamentals are getting better or worse. Joseph Piotroski built it in 2000 for the Journal of Accounting Research as a deliberately crude screen: nine tests drawn from the financial statements, each worth one point, summed into a score from 0 to 9. There are no coefficients and nothing to estimate, which is the point: a binary scorecard cannot overfit a sample, and you can apply it to a company no analyst follows.
Read the score as a measure of direction, not level. Six of the nine signals compare this year with last year. A company earning a modest 4% return on assets can score 9 if every ratio is improving, while a company earning 20% can score 3 if margins, liquidity and turnover are all sliding. That is the opposite of what most ratio screens do, and it is why the F-Score adds information a profitability screen does not.
Piotroski applied it inside the highest book-to-market quintile — statistically cheap stocks, a group holding both genuine bargains and broken businesses. He reported that selecting the financially strong ones raised the mean annual return to that value strategy by at least 7.5 percentage points, and that a portfolio long the 8-and-9 firms and short the 0-and-1 firms returned about 23% a year over 1976 to 1996, with the gains concentrated in small and mid-cap firms that few analysts covered. It is not a distress model: the Altman Z-Score asks whether a company resembles firms that went bankrupt, while the F-Score asks only whether this year's statements look better than last year's.
The nine signals and the logic behind each family
The signals fall into three families, and the split matters: a 6 built entirely from profitability reads very differently from a 6 built on funding.
Profitability, four points. The first two ask whether return on assets and operating cash flow are positive — the minimum evidence that the business generates something. The third asks whether return on assets improved. The fourth is the important one: it awards a point when operating cash flow, scaled by assets, exceeds return on assets. Because both use the same denominator, that reduces to cash flow exceeds reported profit. Earnings that run ahead of cash are accrual-heavy, and accruals reverse. Piotroski singled this signal out as the earnings-quality check, and the accruals ratio measures the same idea on a continuous scale.
Leverage, liquidity and source of funds, three points. A point for long-term debt falling as a share of average assets, a point for the current ratio rising, and a point for not having issued common equity. The last is a revealed-preference test rather than a ratio: a firm that sells stock is telling you it could not fund itself from operations or from the debt market on acceptable terms.
Operating efficiency, two points. A point for a rising gross margin, which captures pricing power before overhead choices intervene, and a point for rising asset turnover. Together they decompose an improvement in return on assets into its two economic sources, as a DuPont decomposition does.
The scaling conventions are not decoration. Return on assets, operating cash flow and asset turnover are divided by total assets at the beginning of the year, so a mid-year acquisition does not flatter the denominator with assets that had no time to earn. Leverage is divided by average total assets, because a debt ratio measured against a fast-moving asset base would move for reasons unrelated to borrowing.
Worked example: scoring a company that improves on eight of nine tests
Take a manufacturer with these figures. Current year: net income $38.6m, operating cash flow $62.4m, revenue $884m, gross profit $297m, total assets $745m, long-term debt $172m, current ratio 1.78, shares outstanding 41.4m. Prior year: net income $31.5m, revenue $812m, gross profit $268m, total assets $700m, long-term debt $190m, current ratio 1.62, shares outstanding 42.0m. Total assets two years ago were $640m.
- Return on assets. $38.6m ÷ $700m (the opening balance sheet) = 5.51%. Positive, so 1 point.
- Operating cash flow. $62.4m is positive: 1 point. Scaled by opening assets it is 8.91%.
- Change in return on assets. Last year $31.5m ÷ $640m = 4.92%, so the change is +0.59 points. 1 point.
- Accrual test. 8.91% of assets in cash against 5.51% in profit, or $62.4m against $38.6m. 1 point.
- Leverage. $172m ÷ (($745m + $700m) ÷ 2) = 23.81%, against $190m ÷ (($700m + $640m) ÷ 2) = 28.36%. It fell: 1 point.
- Current ratio. 1.78 against 1.62: 1 point.
- Equity issuance. 41.4m shares against 42.0m — the count fell, so no stock was sold. 1 point.
- Gross margin. $297m ÷ $884m = 33.60% against $268m ÷ $812m = 33.00%. 1 point.
- Asset turnover. $884m ÷ $700m = 1.2629 against $812m ÷ $640m = 1.2688. Turnover fell: no point.
The score is 8 of 9: profitability 4, leverage and funding 3, efficiency 1. The single failure is the instructive part. Revenue grew 8.9% while assets grew 9.4%, so the company is adding capacity slightly faster than sales. That is not alarming on its own, but it is the question to put to management — and pointing at the one number that moved the wrong way is what a nine-point tally is for.
The nine signals, their definitions and their denominators
| # | Signal | Family | Point awarded when | Scaled by |
|---|---|---|---|---|
| 1 | ROA | Profitability | Net income is positive | Opening total assets |
| 2 | CFO | Profitability | Operating cash flow is positive | Opening total assets |
| 3 | ΔROA | Profitability | ROA is higher than last year | Opening total assets, each year |
| 4 | Accruals | Profitability | CFO exceeds net income | Opening total assets |
| 5 | ΔLeverage | Leverage / liquidity | Long-term debt ratio fell | Average total assets |
| 6 | ΔLiquidity | Leverage / liquidity | Current ratio rose | Current liabilities |
| 7 | Equity issuance | Leverage / liquidity | No common stock was sold | Not a ratio |
| 8 | ΔGross margin | Efficiency | Gross margin rose | Revenue |
| 9 | ΔAsset turnover | Efficiency | Asset turnover rose | Opening total assets |
Maximum score 9: four profitability points, three leverage and liquidity points, two efficiency points.
How to read the score: bands, context and direction
Treat 8 and 9 as the high-signal group and 0 and 1 as the low-signal group, because those are the tiers Piotroski actually tested. Everything between is a crowded middle where the total carries little information on its own, and the useful output there is which signals failed rather than the sum.
The score was validated inside a value screen. Piotroski's sample was the cheapest quintile by book-to-market, so the documented outperformance belongs to the combination of a low price and a high score. Use the F-Score to choose among candidates you already consider cheap, not to justify a price.
It is a one-year snapshot of change. An easy prior year makes this year look strong, so a company recovering from a disastrous twelve months often scores 8 or 9 on mean reversion alone. Compute it for three or four consecutive years: 7, 8, 8 is a genuinely improving business, while 2, 3, 9 is usually a rebound off a low base.
Some sectors break individual signals. Banks and insurers have no meaningful gross margin or asset turnover, so two of nine signals are noise. Regulated utilities raise debt and equity to fund approved capital programmes, which costs them signals 5 and 7 for reasons that carry no negative information.
Where the score is weak, read which family failed. Losing all four profitability points is an earnings problem; losing 5 and 7 means the firm is funding itself externally; losing 8 and 9 means it is being squeezed on price or is over-invested. Pair the result with the return on assets calculator for the level rather than the direction, and with debt to assets for the absolute leverage a change signal hides.
Where implementations quietly disagree
Three definitional choices split published implementations, and each moves the score by a point on real companies. First, the debt-free firm: Piotroski awards signal 5 only when the leverage ratio falls, so a company with no long-term debt in either year sees an unchanged ratio of zero and earns nothing. This calculator follows that strict reading and flags it; several screeners award the point instead. Second, the equity-issuance test is defined on whether common stock was sold, not on the share count, so option exercises, stock dividends and splits can fail a share-count proxy without any capital being raised. Third, long-term debt includes its current portion; excluding that makes leverage look as though it is falling whenever maturities approach.
Mistakes that corrupt an F-Score
- Scaling return on assets by closing or average assets. Piotroski uses assets at the beginning of the year. The closing figure understates ROA for any growing company and can flip signal 3.
- Using operating income instead of net income. The definition is net income before extraordinary items; EBIT breaks the comparison with cash flow in signal 4.
- Skipping the year before last. The prior-year ROA and turnover need their own opening balance sheet, not last year's closing one.
- Ignoring restatements. Take both years from the same filing so a discontinued operation or a new standard has been applied to each.
- Reading a share-count increase as an equity raise. Splits, stock dividends and option exercises raise the count without raising capital.
- Applying the score to banks, insurers or REITs. Gross margin and asset turnover are meaningless for them, so the score tops out at 7 for structural reasons.
- Scoring a single year. One observation cannot separate a genuine turn from a rebound off a bad comparative.
Where the F-Score fits among quality and distress models
The F-Score sits between valuation screens and forensic models, and the three answer different questions.
Distress models ask whether a firm will fail. The Altman Z-Score weights five ratios into a discriminant score with published cut-offs; Ohlson's O-Score produces an actual bankruptcy probability. Both look at level, not change, so both will flag a chronically weak company that the F-Score happily scores 8 because it improved.
Manipulation models ask whether the numbers are honest. The Beneish M-Score uses eight variables to estimate the likelihood that earnings have been manipulated, overlapping with signal 4 without duplicating it. A company that passes eight signals while failing the accrual test deserves that treatment.
Run the F-Score as a second-stage filter: screen on price first, score the survivors, then read the failing signals rather than the total. Follow up on cash conversion with the accruals ratio, and on funding capacity with the sustainable growth rate, which tells you whether the growth these statements imply can be financed without the equity issue that costs signal 7.
Key terms
- F-Score
- The sum of nine binary fundamental signals defined by Piotroski (2000), running from 0 to 9. Higher means broader year-on-year improvement.
- High-signal / low-signal firm
- Piotroski's labels for firms scoring 8–9 and 0–1. These are the only two tiers the original study tested as portfolios.
- Accrual signal
- Signal 4: a point when operating cash flow exceeds net income, testing whether reported profit is backed by cash rather than by estimates.