Relative strength

Level 6 — strategy families.

The claim

Out of a group, the few that rose the most keep outrunning the rest.

Every other family in this section judges one instrument on its own history. This one is different in kind: it ranks a set and holds the top few, whatever their individual charts look like.

The distinction matters more than it sounds. A rule asking "has this gone up?" can be satisfied by a rising market lifting everything. A rule asking "has this gone up more than its peers?" cannot — in a market where everything rose 20%, the ranking still separates the field. You are betting on relative performance, and taking a position on the market itself only incidentally.

Why anyone believes it

Cross-sectional momentum is among the most replicated results in finance. Ranking a universe by trailing return and holding the top decile has been documented across decades, countries and asset classes. Whatever else is arguable, the pattern is not obscure.

Money chases relative performance. Funds are measured against peers and benchmarks, so capital flows toward what is outperforming. The flow is the mechanism, and it operates on relative rank rather than absolute return.

Ranking cancels the market. A shared move affects every member equally and drops out of the comparison. What is left is the differences between names, which is a cleaner signal than any single name's path.

Say the cost

This family's documented edge came with brutal, occasional reversals — "momentum crashes" that arrive at turning points and undo years in weeks. The long-run record is strong and the path is not something most accounts can hold through. Both halves are the finding.

What it looks like

No chart. That is the honest answer, and it is why this is the hardest family to learn visually: the signal is a table, sorted.

Rank	Instrument	Return over lookback	Held?
1	one of the set	+31%	yes
2	another	+24%	yes
3	another	+18%	no, with hold_top: 2
…	…	…	no

cross_sectional_momentum takes two numbers:

Parameter	Means
lookback	how many bars the ranking return is measured over
hold_top	how many of the top-ranked names to hold, and nothing else

Rebalancing follows the ranking: a name that falls out of the top hold_top is sold and the new entrant bought.

When it fails

The reversals are the whole risk. At a market turn, yesterday's leaders become the worst performers, and a rule holding exactly those names takes the full hit simultaneously across every position. Your diversification is nominal — the positions were selected by the same criterion, so they fail together.

hold_top too small is a coin toss. Holding the top 2 of a set of 5 is a concentrated bet on two names. Arvo refuses the degenerate cases outright, which is worth knowing before you try them.

Turnover is high. Rankings churn, and every change is a sale and a purchase, both paying costs. A short lookback makes this much worse.

And the trap that sinks most attempts at this family: survivorship. If your set is "today's index members", you have selected on survival, and every name that failed is absent. The rule will look excellent, and the result is worthless. This family is more exposed to that trap than any other, because a set is exactly the thing people assemble carelessly.

Arvo cannot detect this for you, and says so

Deflation corrects for searching. A universe of today's index members involved no search, so every gate passes such a result honestly. This is why a universe in Arvo refuses to exist without a written reason for its membership, and a file without one is rejected. That written reason is the only available defence, and it is yours to write. "The S&P 100 as constituted in January 2024, before the test window" is a reason. "Large US tech names" is not.

Test it

This rule cannot be run as an ordinary study. Arvo refuses it:

Cross-sectional momentum ranks instruments against each other and needs more than one; run it as a book

And a book of two or three is refused as well:

Cross-sectional momentum ranks instruments against each other; 4 is the fewest a ranking says anything about, and this has 3

So: define a universe with its stated reason, then run a book — many instruments, one account, one gate. A ruleset over the shipped defaults:

{
  "name": "leaders",
  "label": "Cross-sectional momentum",
  "premise": "The strongest few of a stated set keep outrunning it, net of the turnover.",
  "interval": { "step": 1, "unit": "day" },
  "kind": {
    "kind": "grid",
    "rule": "cross_sectional_momentum",
    "fixed": { "trade_size": 10.0 },
    "axes": { "lookback": [20.0, 60.0, 120.0], "hold_top": [2.0, 3.0] }
  }
}

Six configurations, so the winner is judged as the best of six.

What to read:

1. Your universe's reason, before any number. If you cannot defend the membership in a sentence that does not reference performance, stop — the result cannot mean anything, and no part of Arvo will catch it.
2. Which members were silent. A book reports them. A result driven by two of twenty names is a concentrated bet wearing a portfolio's clothes.
3. The regime is the book's, not any member's. For a book, Arvo labels the regime from the benchmark holding every member — so a name that ranged while the book trended is not visible. A stated limit, not a hidden one.
4. The drawdown, hard. This family's failure is a fast simultaneous one. Check it against the drawdown table.

Watch

• (to be chosen) — cross-sectional versus time-series momentum, side by side.
• (to be chosen) — a momentum crash, with the leaders before and after.

Next

Opening range — the first minutes of a session.