Investable Universe: The List Behind Every Stock Screen

Two traders run what they call the same momentum screen on Monday morning. Both rank US equities by six-month price change, both take the top twenty, and both trust the result. Their lists share maybe eleven names. One trader kept every ticker that reported a price. The other had already dropped anything under $5 and under a million dollars in average daily turnover, so half of the first trader’s hot names never entered the comparison. Same formula, two answers. The difference is the investable universe each screen was allowed to see, and most screeners never show it to you.

That gap traces back to the list each screen started with, not to the ranking math. The math ran perfectly on both. It just ran on different populations, and the population is a choice made long before the first score is computed.

From a market label to an investable universe

“US equities” sounds like a set you could point to. In practice it’s a label, and a label is not something a computer can rank. Before any score exists, someone has to turn that phrase into rules. Which exchanges count. Which security types are in: common shares, sure, but what about ADRs, REITs, closed-end funds, SPACs, and units. What minimum price qualifies, how much liquidity a name needs, whether a usable price history even exists for the window, what market-cap floor applies, how a foreign-domiciled company that lists in New York gets classified, how long a stock must have traded before it’s eligible, and what happens to a name after it delists. Every one of those is a decision. Together they are the investable universe.

When I build a universe for a swing screen, I usually start by cutting anything trading under about $5 and under roughly a million dollars in average daily dollar volume, then I require at least a year of trading history so the ranking window has real data to work with. Those are my working defaults, not numbers anyone should copy. I give the exact figures only to make the point concrete: the moment I move the price floor from $5 to $1, or drop the history requirement from twelve months to three, the eligible list changes, and every ranking built on it changes with it.

The common misread is treating that starting list as neutral and self-evident, as if the universe were handed down by the market rather than picked by whoever ran the screen. A screen begins with a filtered snapshot of the market, one that carries the filter’s assumptions, and the ranking inherits those assumptions whether or not anyone wrote them down.

How eligibility reshapes a cross-sectional rank

A cross-sectional ranking compares every security against its peers on the same date, then sorts. Percentile ranks, factor averages, sector weights, the apparent spread of outcomes: all of them are measured relative to whoever else is in the pool. Change the pool and you change the answer, even when you never touch the formula.

Say a name sits at the 90th percentile of a 500-stock momentum rank. Add 1,500 recently listed, high-volatility microcaps that have screamed higher off tiny bases, and that same steady name can slide toward the 70th percentile without moving a cent. Its raw momentum held exactly where it was. The crowd it’s ranked against is what changed. The top decile now fills with fragile names, the sector mix tilts toward wherever the new listings cluster, and any factor average you report drifts toward the newcomers. A multi-factor composite screen layered on top compounds the problem rather than rescuing you, because every factor in the composite gets percentile-ranked against the same altered pool.

Investable, benchmark, and research universes

Three different universes get used in the same breath, and they answer three different questions. The investable universe is the set I could actually trade at my size, after liquidity and price filters, without pushing the market against myself. The benchmark universe is the yardstick, the set whose return I’m measuring against, often an index. The research universe is the full set of names I let a hypothesis see while I’m testing whether an effect is even real.

These pull apart fast. A microcap effect can be strong in a research universe and untradeable in my investable universe, because the names that produce the effect are exactly the ones I can’t fill. An index makes a convenient benchmark, but an index constituent list isn’t automatically the right research universe. Indexes have their own eligibility committees, their own float and liquidity rules, and index reconstitution that adds and drops names on a schedule set for reasons that have nothing to do with your study. Borrow the index as your research set and you’ve inherited someone else’s universe design without meaning to.

When the reconstructed history lies

Backtests are where universe design does its quietest damage. To test a rule over ten years, you need to know which securities were eligible on each past date. If you rebuild that membership from today’s list, you’ve handed the past information it never had. The names that went to zero, got acquired, or delisted have already fallen out of today’s list, so a universe reconstructed from current members silently deletes the failures before the test even runs.

Keep two ideas apart here, because they get blurred constantly. Survivorship bias in a backtest is about what remains in the historical data after the losers have been scrubbed out. Universe design is the broader, earlier decision about what was ever eligible to be compared. Survivorship is one way a badly built universe goes wrong; it isn’t the whole story. You can keep every delisted name, beat survivorship bias cleanly, and still mis-specify the universe by applying a market-cap floor as of today rather than as of each historical date. A point-in-time universe stores what was eligible on each date and never lets a later fact leak backward into the test.

Fixed membership versus a refreshed universe

Two honest studies can disagree because one holds the universe fixed and the other refreshes it. A fixed universe locks the eligible list on a start date and follows those exact names forward. That answers a specific question: how did this particular set of companies behave from here. A refreshed universe re-runs the eligibility rules every period and lets membership turn over. That answers a different question: what would the rule have selected, period after period, as companies grew into the filters and dropped out of them.

William O’Neil built CANSLIM around screening the whole market for leaders over and over, which is a refreshed universe by design: the leaders change, so the eligible list has to. Refreshing isn’t free. Every reconstitution of the list creates turnover, and a rule that looks clean on a fixed universe can bleed its edge to costs once membership churns each quarter. Neither version is more correct. They’re answers to different questions, and the mistake is quoting a result from one as if it settled the other.

One rank, two lists

Here’s the effect in miniature, with illustrative numbers so the mechanism is visible. Rank a universe of 400 large and mid-cap names by twelve-month momentum and note the top ten. Now widen the same screen to include recently listed shares and microcaps below your old liquidity floor, so the pool jumps to 1,600 names. When I’ve run this kind of before-and-after on my own watchlists, the top ten barely survives the change: several fresh listings with three months of history and violent percentage moves leapfrog the seasoned names. The shape of the list shifts too. The effective number of holdings needed to express that top decile changes, the sector mix rotates toward wherever the new names cluster, and the reported average momentum of the leaders climbs for a reason that has nothing to do with any company getting stronger.

I’m not going to hand you a cutoff and call it correct, because there isn’t one. A $5 floor, a $1 floor, a 500-name cap, a full-market sweep: each is defensible for some question and poor for another. The exercise makes one thing visible. The filter produced the ranking as surely as the formula did, so two people can run identical scoring code and still defend contradictory lists with a straight face.

Ask what list it started from

Transparent eligibility rules buy you interpretability. When the universe is written down, you can see why a name is present, reproduce the screen, and argue about the filter instead of mistaking it for nature. That’s worth a great deal. It doesn’t buy you truth. A clean, documented universe still carries model risk, still breaks when the market regime shifts, still suffers from a bad data feed or a mislabeled delisting, and can still exclude the very names that would have mattered most. Writing the rules down makes a screen honest, not right.

So the first question to ask about any ranking, backtest, or leaderboard is the dull one: what list did this start from. Who was eligible, on what date, under which price, liquidity, and history rules, and what happened to the names that died along the way. Get that answer and the scores start to mean something. Skip it and you’re reading percentile ranks off a pool you never inspected. Learn the pattern. Ride the trend. Keep the gains.

Educational content only. Not investment advice. Trading involves risk. You are responsible for your decisions.