2015issue C1234-37

Evaluating rank rotation after a persistence screen

A historical workflow first keeps names whose annual switch-rate stays below 63, then tests a one-name rank-rotation overlay by adding names one at a time. Occupied-day share and the overlay annual figure are read together so coverage can be compared with dilution.

A name is treated as persistent only when its annual switch-rate stays below 63. In the ten-name window from 30 June 1994 through 1 July 2014, every listed name met that screen.
Rank-rotation holds only the currently strongest eligible name and stays flat when none qualifies. The one-rank rule was out of the market about half the time, so its occupied-period rate of change exceeded its calendar annual figure.
In the greedy add-one sequence, days spent flat fell from 53.3 percent with one name to 9.8 percent with ten names, while the overlay annual figure rose through six names and then declined. Every ten-name path ended at the same 54 percent annual figure.
The two-name overlay accepts a name's next return only through a rule-based-entry gate: that name must be the current leader, and its prior return must clear a near-zero threshold. Apparent best roster size still depends on the add-one path.

Entries in this reading3 entries

Rule-based entry5 readings Momentum strategy16 readings Current trackRank rotation50 readings

Two checks in one procedure

This article treats a persistence screen and a rank-rotation overlay as a single, testable procedure. Editorial framing: first confirm short-term memory with a switch-rate cutoff, then measure whether adding names one at a time improves occupied-day coverage more than it dilutes the overlay figure. Rank-rotation holds only the currently strongest eligible name and stays flat when no name qualifies. The overlay is a momentum-strategy: it conditions participation on recent continuation rather than on a static basket.

Confirm persistence with switch-rate

A name is treated as persistent only when its annual switch count stays below 63. Switch-rate is the annual count of long-to-flat or name-to-name changes used as that screen. The ten-name study window runs from 30 June 1994 through 1 July 2014. Across those ten names, every listed switch-rate is below 63, so each name meets the persistence screen.

Start from the one-rank overlay

One-rank is a single-name persistence rule that is long only after a non-negative prior return and is otherwise out of the market. The one-rank rule is described as being out of the market about half the time, so its occupied-period rate of change is larger than its calendar annual figure. Occupied-day share is the fraction of calendar days the composite is long any eligible name rather than flat.

Dual overlays and the entry gate

Pairing the ten names produces 45 dual combinations, and the strongest dual overlay in that grid is RES with FAST. The two-name overlay accepts a name's next return only when that name currently has the larger prior return and that prior return clears a near-zero threshold. That if-and check is the rule-based-entry gate. Equity and max-drawdown series are then built from the summed accepted returns.

Adding names one at a time

In the greedy add-one sequence, the overlay annual figure rises through six names and then declines toward the full ten-name mix. As names are added in that sequence, days spent flat fall from 53.3 percent with one name to 9.8 percent with ten names, while up-day and down-day shares both rise. Because the overlay depends on the set of names rather than insertion order, every ten-name path ends at the same 54 percent annual figure.

Roster size follows the path

The greedy path peaks at six names, the one-rank descending path at five, an alphabetical path at eight, and the reverse one-rank path only at ten, so the apparent best roster size is path-dependent.

Overlay annual return as names are added one at a time

The add-one overlay is path-dependent: choosing the next name by ex-post contribution lifts annual return to 71.5 percent at six names, then dilution pulls it back to 54 percent at ten. Adding the same persistence-screened names in descending one-rank order peaks earlier, at five names and 59 percent, and still finishes at 54 percent. Along the optimized path, days out of the market fall from 53.3 percent to 9.8 percent, so occupied-day coverage and return dilution have to be read together. The plotted annual figures are taken from the source combination tables, not traced from the printed curves.Ten persistence-screened US stocks led by RES · 20 years, June 1994–July 2014 · 1994-06-30T00:00:00.000Z to 2014-07-01T00:00:00.000Z

The source treats the upper path as in-sample cherry-picking and offers the one-rank order as the more realistic insertion rule. Both rosters keep only names with fewer than 63 switches per year. Study window 30 June 1994 to 1 July 2014. Alphabetical and reverse-rank paths appear on the printed plot but were not tabled, so they are omitted.

Educational research material, not investment advice. Historical source context does not establish present-day performance.

25 of 38 in the Rank rotation track

201610-10 pp.Next on Rank rotationEvaluating an annual valuation rank rotationABR is the signed gap between a stock's earnings-yield and a required-return built from a 10-year government bond yield plus an equity risk-premium estimate.

All readings on this track · 38 readings

By fieldTrend and mean reversion10 readings By year201516 readings

All 50 readings tagged Rank rotation

Also on Rank rotation5 readings

2002Six-week reversal candles, next-week entry, and a break-even stopRule-based entry 2005Building entry rules with ratchet volatility stopsRule-based entry 2005Pricing entries, stops and exits in range unitsRule-based entry 2006Permission and fill gates for mechanical systemsRule-based entry 1992Currency rank rotation and intermarket timingMomentum strategy