Segmentation study · recommended version · screener snapshot 22 Sep 2026

Nine kinds of stock

2,000 listed stocks in 9 groups and 19 subgroups, based on what the businesses look like on valuation, profitability, growth, balance sheet, dividends and beta. Size and price momentum describe the segments but were not used to form them. This is the recommended version, chosen over the 6-group benchmark and the 12-group variant.

2,000stocks segmented

9 / 19groups / subgroups

0.80agreement across 5 seeds

73smallest group in any run

0.201sector × segment co-movement

Why nine

No number of groups is statistically "best". Anywhere from 9 to 13 groups is equally repeatable (0.80–0.82 agreement across 5 seeds). What separates them is structure:

Nine is the only option that never throws off a sliver. Its smallest group is 73 stocks in every run. At 10–13 groups, some runs produce a group of just 2–14 stocks.
It has the best separation of any option from 8 to 13 (silhouette 0.126, Calinski-Harabasz 215).
It gives the cleanest two levels. Its 19 subgroups recover almost all of the 12-group version's distinctions, such as blue chips vs levered buyback franchises and growth leaders vs net-cash compounders, one level down.

The segments

Medians for each group. Colours show the benchmark group each one mostly comes from. Subgroups list their largest companies.

Segment fingerprints

Average score of each segment against the whole universe. Scores are rank-based z-scores, where 0 is the universe median and ±1 is about the 16th/84th percentile. The last block of columns describes the segments but was not used to form them.

Segments cut across sectors

Share of each group's stocks in each sector; rows add up to 100%. Technology is spread across three groups, and financials split into banks and insurers. The groups say something the sector labels don't.

How the 9 relate to the benchmark 6

Number of stocks from each benchmark group (columns) that land in each new group (rows). The benchmark's catch-all "Market core" splits into mature franchises and low-margin growers, and its unprofitable growth splits into pre-profit growers and cash burners.

Do segments explain co-movement?

Average pairwise correlation of weekly returns within groups minus between groups, over the last 104 weeks, for the 690 stocks with price history. A higher gap means the grouping captures stocks that move together. Teal bars are this version.

How to use this: as a second dimension next to sector. On their own, segments explain less co-movement than sectors. Combined with sector they add real information: a random split of each sector adds nothing. The 12-group version scores a little higher (0.221), partly because it has more, smaller cells.

How it was built

17 features in 5 equally weighted themes (valuation, profitability, growth, balance sheet, income & risk). Ratios are rebuilt from line items, so earnings yield exists for loss-makers too.
Tie-aware rank→normal transform, then PCA to 9 components (88% of variance).
Banks, brokers and asset managers (240 stocks) get neutral free-cash-flow and balance-sheet scores. Their FCF swings with loan books (JPM's FCF margin is −81%), and their "debt" is deposits.
Consensus clustering. k-means runs 60 times on 80% subsamples. The groups come from how often each pair of stocks lands together, which stays repeatable where plain k-means does not.
Choosing k: the most granular k from 8 to 13 whose runs agree at least 0.78 across 5 seeds and whose smallest group stays at 40+ stocks in every run. Only 9 passes.
Subgroups come from k-means within each group, used only where the split is repeatable (stability ≥ 0.60) and every part has 40+ stocks.

Choosing the number of groups

Agreement is the mean (and worst) ARI between the 10 pairs of five independently seeded runs. "Smallest" is the smallest group seen in any of the five runs.

Versions compared

Version	Groups / subs	Smallest	Segment gap	Sector × seg.
Benchmark (k-means)	6 / 13	257	0.064	0.182
Recommended (this page)	9 / 19	73	0.065	0.201
Finer cut	12 / 24	11*	0.077	0.221

* 48 in the published run, but some seeds produce an 11-stock sliver.

Where the 12-group distinctions went

Here (subgroups)	12-group version
1.1 · 1.2	Mature blue chips · Levered buyback franchises
2.1 · 2.2	Growth leaders · Net-cash compounders
3.2 and 9	Broken growth, split between the two
Groups 4–8	Same as the 12-group version's matching groups

Find a stock

Search by ticker, name, sector or industry.

Caveats

Most subgroup splits are solid (stability 0.79–0.92), including groups 1, 2, 3, 5, 6 and 8. The low-margin growers' split is softer (0.72). Yield & real assets has no repeatable split. Cash burners & broken growth (92 stocks) is too small to split into two parts of 40+.
"Banks & lenders" is 29% non-financial. Companies such as BHP, CME, BAM and VICI share the banks' profile of high margins, high yield and low beta. Most of them sit in subgroup 5.3.
Yield & real assets includes dividend pharma. ABBV, MRK, PFE and GILD sit here beside utilities and REITs, because of high dividends and debt with flat growth.
Broken growth has no group of its own. Former growth stories such as INTC, COIN and WBD are split between the break-even cyclicals (3.2) and the cash burners (9). The 12-group version keeps them separate if that story matters.
The co-movement test is large-cap biased (690 of 2,000 stocks have return history). This is one snapshot, and it is descriptive, not predictive.