The gate exit rule

The criterion that buys a company is the criterion that sells it. A name is held only while it still clears all six quality gates, and it is dropped the quarter it stops. The book is rebuilt from scratch every quarter, so a holding persists only by re-qualifying on its own merits — not on how far it has already run.¹

What it produced

1. The beta cap. The book also requires a constituent’s trailing one-year beta to be ≤ 1.30, measured only on prices before formation (the same rules, no beta cap row above is the book without it). That single constraint pulls portfolio beta from 1.34 to 1.05 and holds it there in both halves, and volatility falls in both while Sharpe is never worse. The return effect is not robust — +6.2pp in the first half and −5.3pp in the second, a sign flip that should be ignored — so the cap is a risk control, not a return engine. It is a footnote to the exit rule, not the point of it.

Putting a factor model into the mix

Can re-weighting the score toward what the market actually pays for beat the score itself? Tested honestly — no, and that is the useful answer.

An elastic-net characteristic model on the same survivorship-free panels — every fundamental as of each formation, set against the next quarter’s beta- and sector-neutral return — finds two factors carry the cross-section: quality and cash generation. It holds up out of sample: a weighting learned on 2014–2022 predicted 2023–2026 with an information coefficient of +0.048 (t = 3.9).

So the construction is held identical — six gates, beta ≤ 1.30, top 25, rebuilt quarterly — and only the ranking changes: from the composite score to a factor-weighted score whose weights are fit strictly on data before each quarter, never after. If the model knew something the score did not, the tilted book would pull ahead.

It does not. Out of sample the tilted book returns essentially what the score already does — +0.0pp a quarter, t = 0.25, ahead in just 10 of 42 quarters. The two curves sit on top of one another, and both clear the S&P by the same margin.

Why that is the right result. The composite score already blends quality and cash; the model, fit blind, rediscovers the same two factors — so it validates the construction rather than beating it. Re-weighting the book for a +0.0pp, t = 0.25 effect would be fitting noise, so the holdings are left exactly as they are. The model earns its keep as a standing check on what the market is paying for — a way to notice if that ever stops being quality and cash — not as an override.

Trading activity

The book is rebuilt from scratch every quarter, so a name persists only by re-qualifying on its own merits. Fewer than 25 names appear whenever the beta screen leaves a thinner eligible set than the gates alone would.

Every buy and sell

Step through the formations. Buys are companies entering the book that quarter; sells are companies that were held and no longer qualify. The table beneath is the resulting portfolio, with the trailing beta each name was screened on. Mean name beta is the average of those trailing figures, which is not the same thing as the book’s realised beta of 1.05 — the first is measured before formation, the second from what actually happened after.

Book value against the index

Both start at £100,000 on 31 December 2013.