← Quantery Blog

Should a stock screen use levels or changes? | Quantery Blog

October 23, 2026 · 8 min readthesis-craftstock-screeningfundamentals

A stock screen should use a level when the thesis requires a condition to be good now, and a change when the thesis cares about direction. Use both when strength and improvement are separate parts of the claim. Keep them as separate criteria so you can see which one admitted a company and which one rejected it.

A high operating margin can be shrinking. A low margin can be recovering. Neither fact cancels the other, and folding both into one number loses the distinction. Write the level and the change separately, decide which one is essential, then test whether the second signal adds useful information.

What is the difference between a level and a change?

A level is the value of a metric at a chosen point or over a chosen window. Current ratio in the newest filing is a level. Trailing free-cash-flow margin is a level too, even though it combines several periods. It answers: where is the business now?

A change compares that metric with an earlier observation. The current ratio rising since the same quarter last year is a change. Operating margin expanding year over year is a change. It answers: which way is the business moving?

The distinction sounds basic. It gets muddled inside a composite score because the two often share inputs. Current operating margin and its year-over-year change both use revenue and operating income. That doesn't make them duplicate signals. One rewards the economics already in place. The other rewards the direction of travel.

Put the possibilities side by side:

• Strong and improving: the business clears the present-tense standard and its latest comparison moved the right way.
• Strong and weakening: the level still clears the bar, but the direction warns that the advantage may be narrowing.
• Weak and improving: the operation hasn't reached the standard, though the change can identify a recovery candidate.
• Weak and weakening: neither version of the claim has support.

A single cutoff sees only two buckets. A single change signal also sees two buckets. The paired view preserves four economically different states.

Microsoft shows why one number isn't enough

Microsoft's fiscal 2023 filing gives a clean example of a strong level with a weaker change. The company reported fiscal 2023 revenue of $211.915 billion and operating income of $88.523 billion. Those figures produce an operating margin of about 41.8 percent. The same filing reports fiscal 2022 revenue of $198.270 billion and operating income of $83.383 billion, which produce a margin of about 42.1 percent (Microsoft 2023 Form 10-K).

Operating income rose, yet operating margin slipped by about 0.3 percentage points (Microsoft 2023 Form 10-K). A level rule asking for a healthy operating margin would read that filing as strong. A change rule asking for margin expansion would read the latest comparison as weaker. Both readings follow from the same document.

Go back one filing and the direction changes. Microsoft's fiscal 2022 revenue was $198.270 billion against $168.088 billion in fiscal 2021, while operating income was $83.383 billion against $69.916 billion. The resulting operating margins were about 42.1 percent and 41.6 percent, so the level was high and the margin had expanded by about 0.5 percentage points (Microsoft 2022 Form 10-K).

This is a demonstration of state. The level stayed high across both filings. The change flipped sign. A thesis that treats those observations as interchangeable can't tell you which claim it is making.

The same split applies elsewhere. Positive return on assets measures profitability now; improvement in return on assets measures progress. A manageable debt load measures current balance-sheet pressure; falling leverage measures repair. Positive operating cash flow and an increase in operating cash flow belong to different questions too. The Piotroski F-score guide is useful because its component signals keep several of these distinctions visible instead of hiding them in one ratio.

Should level or change be the gate?

Make the gate follow the sentence you would defend without seeing any results.

If the thesis is durable quality, the level usually carries the must-pass condition. A business with a thin margin doesn't become a high-margin business because the margin improved from a worse starting point. Improvement can earn points or a flag, but it shouldn't rewrite the claim.

If the thesis is operational recovery, change may deserve the larger role. The research question is whether conditions are getting better before the full quality level returns. Requiring an already excellent level can remove the companies whose recovery you meant to study. Even then, add a floor. Improvement from a deeply negative base can remain economically weak.

If the thesis is quality with momentum, require both. That creates a narrower claim: the business is strong and still moving forward. Expect the paired rule to favor a different set of companies from either component alone. That may be exactly what you intended. It may also select one industry, one market regime, or companies near a cyclical peak.

Don't let a composite score make this choice accidentally. Two points for level plus two for change allows a weak level to be offset by rapid improvement when the gate uses only a total score. A strict gate says each criterion must earn support. Neither structure is universally better. They encode different claims, as the guide to strict gates and score gates explains.

What does a level-and-change thesis look like?

The shape below is illustrative, and the bundled templates are the reference for exact fields. The thresholds are proposed research parameters. They aren't universal definitions of strong profitability.

meta:
  name: level-and-change
  label: Level and change
  description: Separate current operating quality from year-over-year direction.
  scale: { min: 0, max: 2 }

universe:
  exchanges: [NYSE, NASDAQ, AMEX]
  exclude_sectors: [Financial Services]

quality:
  stale_months: 9
  required:
    revenue_now: no_current_revenue
    operating_income_now: no_current_operating_income

params:
  min_history: 5
  margin_strong: 0.15
  margin_ok: 0.08
  improvement_floor: 0.01

features:
  revenue_now: newest(revenue)
  revenue_prior: lag(revenue, 4)
  operating_income_now: newest(operating_income)
  operating_income_prior: lag(operating_income, 4)
  margin_now: if(revenue_now > 0,
    operating_income_now / revenue_now, null)
  margin_prior: if(revenue_prior > 0,
    operating_income_prior / revenue_prior, null)
  margin_change: if(is_null(margin_now) or is_null(margin_prior),
    null, margin_now - margin_prior)

criteria:
  profitability_level:
    rules:
      - { when: "is_null(margin_now)", score: 0, flag: no_current_margin }
      - { when: "margin_now >= $margin_strong", score: 2 }
      - { when: "margin_now >= $margin_ok", score: 1 }
      - { else: 0 }
  profitability_change:
    rules:
      - { when: "history_periods < $min_history or is_null(margin_change)", score: 0, flag: no_yoy_comparison }
      - { when: "margin_change >= $improvement_floor", score: 2 }
      - { when: "margin_change > 0", score: 1 }
      - { else: 0 }

gate:
  mode: strict

newest() keeps each current-period input tied to the newest filing. lag(field, 4) reaches to the same fiscal quarter in the prior year when the filing series is complete. The history check matters because a positional lag can't prove that missing filings haven't shifted the comparison. Trailing and year-over-year numbers need different period rules, and the thesis should show those rules instead of assuming them.

The margin change uses subtraction, so 0.01 means one percentage point of expansion. Record that unit in the research log. Switching to percentage growth can change the screen while leaving the label untouched.

The example uses a strict gate to express quality with momentum. For durable quality, keep the level criterion in the gate and turn the change into context. For recovery, lower the level floor and make sure the remaining floor still has an economic meaning. The rules are yours, but the label has to match them.

How do you test whether the change signal helps?

Start with the level-only thesis. Add the change criterion without touching the universe, dates, reporting lag, rebalance schedule, exit rule, or benchmark. Compare who qualifies before comparing returns. The membership difference tells you what the change rule bought.

Then inspect the excluded strong businesses. Did the rule remove companies with a tiny one-period contraction? Did it remove capital-heavy firms after a normal investment cycle? Did it concentrate the surviving group in industries where margins were expanding together? A change signal can turn a quality thesis into a cycle-timing thesis without changing its name.

Run the reverse comparison too: change alone against level plus change. That shows whether the level floor removes weak recoveries or strips out most of the improvement signal. If you want the full interaction test, use the four-run method for paired stock signals. It separates the contribution of each rule from the result of requiring both.

Read the backtests as hypothetical and before costs. Compare event timing, drawdowns, and membership stability, then test nearby round thresholds chosen in advance. A one-percentage-point improvement hurdle that works while a flat-margin rule and a two-point hurdle fail has found a narrow boundary. It hasn't found a law.

Keep the losing versions. A level-and-change thesis is easy to tune because every metric offers two dials: the acceptable state and the required direction. Quantery's thesis documentation covers the exact DSL, and each run records the exact thesis version it used. Note which dial changed in your research log. The useful result tells you whether the thesis rewards a good business, an improving business, or the smaller group that is demonstrably both.

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

← All articles