Name the economic quantity
Profitability describes earnings relative to resources or capital; quality is a broader label that can include leverage, earnings stability and other attributes. Calling both a quality factor obscures what a model actually measures. Robert Novy-Marx's original research uses gross profits relative to assets. A return-on-equity screen instead asks how much earnings accrue relative to an equity base, producing different rankings.
The French library provides a useful warning about implementation precision. Its operating-profitability portfolio detail page describes a book-equity denominator, while its variable-definitions page includes minority interest alongside book equity. Do not resolve that difference by silently mixing formulas. Pin the source version and chosen definition; a replication requires clarification of the historical implementation rather than reliance on a familiar factor name.
A small balance sheet can manufacture a large ratio
Take fictional Company C with revenue of $500 million, cost of goods sold of $300 million, assets of $800 million, net income of $40 million and equity of $200 million. Gross profits divided by assets equals 25%; net income divided by equity equals 20%. These are separate descriptors, not two estimates of the same quantity.
Now suppose a hypothetical financing and repurchase transaction leaves earnings and assets unchanged for this simplified comparison but reduces equity to $100 million and increases liabilities by $100 million. Return on equity becomes 40%, while gross profitability remains 25%. The higher equity return does not, by itself, demonstrate improved operating productivity. Actual transactions can also change interest expense and future earnings, which this static example deliberately holds fixed.
Translate statements into stable features
A reproducible implementation records whether assets are beginning-period, ending-period or an average; whether earnings are annual or trailing; and how exceptional items are treated. Keep currency units and reporting periods aligned. Financial firms, insurers and industrial companies may not have comparable revenue and cost structures. A universal formula can be technically computable yet economically uninformative across those businesses.
Review missing expenses separately from genuine zeros. An absent tag is not evidence that a company incurred no expense, and a vendor's standardized field may combine several issuer tags. Build a reconciliation from source statement to feature for selected records. Flag tiny or negative denominators before normalization, and report the excluded share of the universe so a clean result does not conceal selective coverage.
Evaluate the descriptor rather than its label
To study a composite quality score, declare the components and their directions in advance. If profitability, low leverage and stability each receive one-third weight, preserve that identity through evaluation. Trying many component sets and reporting only the strongest return makes the result a selection exercise. Compare the composite with its individual components to understand whether diversification or one dominant input explains the behavior.
Profitability can be persistent without being attractively priced, and a strong historical business can experience a rapid change. Evaluate valuation and concentration as separate exposures rather than promising a defensive outcome. The calculations here establish accounting sensitivity only. They provide no estimate of expected returns, and neither the cited academic work nor a named index guarantees the same effect in another universe.
Inspect the example
| Illustrative balance-sheet state | Gross profits / assets | Net income / equity | Interpretation |
|---|---|---|---|
| Before recapitalization | $200m / $800m = 25% | $40m / $200m = 20% | Initial ratios |
| After simplified recapitalization | $200m / $800m = 25% | $40m / $100m = 40% | ROE doubles without assumed operating improvement |
Sources & reproduction
Primary references support the definitions and method. Numerical examples are illustrative unless a source record is explicitly identified.