CompanyGraph
How a stock screen or score turns selected observations into a decision aid—and why the design of the measurement matters as much as the result.
A score is a model of a question
Every evaluation system begins by choosing a question. Is the task to find financially improving value stocks, identify distress risk, compare profitability, estimate valuation, or rank momentum? The system then selects inputs, defines a period and comparison universe, applies rules or weights, and produces a number, rank, or pass/fail result.
The output is not the company. It is a statement about the company under those definitions. A quality score can describe profitability and leverage while ignoring a patent expiry. A cheapness screen can identify a low multiple while ignoring a collapsing market. A distress model can classify resemblance to a historical group without predicting the date of failure.
Measurement, classification, heuristic
A ratio such as return on invested capital is a measurement built from accounting inputs. A model such as the Altman Z-Score is a classification tool calibrated on a particular population and period. A screen that combines low price-to-book with improving cash flow is a heuristic: it can organize research without claiming a universal causal law.
These types should not be mixed. A measured margin is not a probability of competitive survival. A classification score is not a forecast clock. A heuristic can be useful even when it is not a tested theory, but the user must know whether the evidence supports description, prediction, or only prioritization.
Historical models have a population boundary
Piotroski's F-Score was introduced as a rule-based use of historical financial-statement signals within a value-stock universe. Its original study found that the signals helped separate stronger from weaker firms in that setting; it did not establish that the same thresholds work in every industry, market, or future period. The original Journal of Accounting Research study is therefore part of the model's definition, not merely a citation for the idea.
The same discipline applies to distress models. Inputs such as working capital, retained earnings, leverage, profitability, and market value observe accounting and market conditions. They do not observe lender behavior, refinancing access, hidden liabilities, management decisions, or the exact timing of failure. A score can be recalibrated for another population, but the new validation must be demonstrated rather than assumed.
Aggregation creates both coverage and distortion
Combining several measures can reduce reliance on one noisy signal. It can also hide which condition drives the result. Equal weights imply that each input is equally useful for the question. Thresholds can create discontinuities when a small accounting change moves a company from pass to fail. Percentile rankings provide relative position but can make a weak peer group look strong.
Keep the decomposition visible. If a company scores well because of margin and cash flow but poorly on leverage and customer concentration, the profile is more informative than the total. If two companies receive the same score through different inputs, they do not share the same risk.
Financial statements are necessary but partial
The SEC explains that a Form 10-K contains audited financial statements and management discussion, while also warning investors to read the narrative and risk disclosures. The statements provide structured observations about the past and defined accounting estimates. They do not provide a complete operating map of customers, competitors, product quality, or future capital requirements. The SEC's guide to financial statements describes those reporting boundaries.
Price is a separate dimension
Business quality, financial strength, and market price answer different questions. A high-quality company can be priced for perfection. A distressed company can be cheap enough for a specialized turnaround strategy. A score that omits valuation cannot determine expected return, and a valuation ratio cannot determine whether the business will remain healthy.
Time also matters. A score based on annual statements may lag a product failure, covenant breach, or acquisition. Market-based inputs can move daily and reflect expectations rather than confirmed operating change. Combining them without aligning dates can create a false sense of precision.
How to use evaluation systems responsibly
Before applying a screen, record its purpose, source, period, universe, missing-data rules, weights, and rebalance interval. Test whether the historical result survives reasonable changes to those choices. Compare the output with the company's customers, products, competitive position, capital needs, and management incentives.
Use a high or low score to decide what to investigate, not what to believe. Ask which operating condition would make the score wrong, how quickly that condition could change, and whether the market already reflects the measured information. A method earns trust when its boundaries are explicit and its results remain useful after the assumptions are challenged.