The Low-Vol Sleeve (quarterly + hysteresis) — Canonical Reference

Lifecycle: PERMANENT. A canonical strategy reference in the strategies folder layer — one authoritative page per strategy, never archived while the strategy exists. It does not retire; it is UPDATED, and its §Maintenance says what must be kept in sync. Machine-held by an automated test, which fails the build if this page stops being served or stops being listed in the index.

🟢 STATUS — THE ONE CONSTRUCT IN THIS ESTATE THAT CLEARS THE FUNDABLE BAR

🔴 EVERY RATIO ON THIS PAGE IS A RETURN/VOL RATIO, NOT A SHARPE

The 1.06, the 0.89 hurdle and every capacity figure below are computed as mean/sd × √periods with no risk-free rate subtracted. Read as a textbook Sharpe they read high. But the index hurdle is computed on the IDENTICAL basis, so every relative claim here — including the fundability verdict, which is entirely a claim about clearing that hurdle — holds exactly as written. the desk's ruling : relabel the vocabulary, change no numbers.

One-line definition: a long-only, equal-weight book of the calmest liquid Indian stocks — enter the bottom-20% by trailing volatility, hold a name until it leaves the wider bottom-40% band, re-select quarterly — whose low turnover is the entire reason it survives realistic trading costs at institutional size.

Where this stands — fundable on the recorded evidence, and NOT live. This is the one book in this estate that still clears the index's return-per-unit-of-risk after realistic trading costs at institutional size, and its worst fall stays around −21% where the index falls −30%. That is not the same as running it. There is no live book, no published holdings and no money in it: the evidence is a back-test with an honest cost model, and the next step on the record is a forward test that watches it quarterly before any capital. Nothing here is investment advice.

1. What it is — and what it is NOT

It is the estate's answer to a question every other strategy here has failed: is there anything we could actually fund? The standing doctrine is that price strength is the only gross forward-return engine and no factor here is a fundable net-of-cost alpha vs the index. This is the single recorded exception, and it does not work by finding a stronger signal. It works by not trading.

It is NOT an EMA crossover result. 🔴 This matters because the module that forward-tests it, the ema crossover forward code, is a crossover harness — the sleeve rides along as the excluded non-crossover COMPARATOR (the role field in that runner) and is judged for reference, never as a crossover finding. Labelling it one would attribute a real result to a falsified family.

It is NOT STEADY-25. See §7 — that confusion is the single most likely misreading of this page.

It is NOT live. No roster is published, no capital is allocated, and its forward window has not been judged.

2. Our variation vs. the standard technique

The low-volatility anomaly is textbook: the calmest stocks have historically delivered index-like returns with materially less drawdown, which is exactly the shape a defensive allocation wants. Two things here are ours, and only one of them is a signal change.

(a) Hysteresis — a hold band wider than the entry band. The textbook form re-selects the bottom quintile every period, so a name oscillating around the cut-off is bought and sold repeatedly for no reason other than measurement noise. Here a name is BOUGHT at bottom-20% and HELD until it leaves bottom-40%. (b) Quarterly, not monthly, re-selection.

Together those cut turnover from 199%/yr to 71%/yr — a 2.8× reduction — and that single change is the whole result. It costs a little frictionless edge (the wider band dilutes the signal: 1.18 → 1.07 R/V before costs) and buys roughly 5× the capacity, while flattening the drawdown profile across every AUM tested.

3. How it works — the ruleset (definitional)

Universe. Liquid NSE names on the shared PIT machinery — bhavcopy EQ+BE+BZ, corporate-action adjusted, split-ratio quarantine, no survivorship bias. Signal. Trailing realised daily volatility per name (sigma_daily), ranked cross-sectionally. Entry. The bottom 20% of that volatility ranking. Hold (the hysteresis). A held name is retained until it leaves the bottom 40% band, then dropped at the next re-selection. Cadence. Quarterly re-selection. Weighting. Equal-weight across holdings. Costs. Standalone numbers are net of a flat 0.15%/side on realised quarterly turnover; the capacity table is re-cut with a participation-cost model (cost_participation.side_costs, Almgren form) at each AUM. Benchmark. Nifty 500 buy-&-hold, return/vol 0.89, on the identical no-rf basis.

4. Status, validation & the fundability verdict

Standalone, net of flat cost: return/vol 1.06· CAGR 15.0%· MaxDD −20.8%. Walk-forward, both halves balanced: h1 1.05 / 13.2% / −15.4· h2 1.08 / 16.6% / −20.8. Correlation to the momentum stack: 0.003 — this is a genuinely independent return stream, not momentum wearing a different label.

The capacity fence is the finding. Net return/vol· CAGR· MaxDD, by assets under management, against the index's 0.85 / 13.4% / −30%:

Median held ADV ₹50.8cr; soft-capacity estimate ₹622cr. The drawdown does not degrade with size — it sits at ~−21% at every AUM, where the monthly predecessor bled to −30%. That stability, not the return, is the reason this is the defensible product: it beats the index on return to about ₹250cr and beats it on drawdown everywhere.

The blend is NOT the product. A 40/60 momentum/low-vol mix scores a higher ratio (1.25) but drags the drawdown to −34.1% and cannot scale — momentum's ₹4.6cr median ADV binds the whole blend to small AUM. Recorded, and rejected as the fundable form.

What "fundable" does and does not mean here. It means: on the recorded evidence, net of a realistic cost model, this clears the index hurdle at sizes that matter. It does not mean deployed, and it does not retire the standing doctrine — the ledger's fundability corollary is narrowed by this result, not overturned. The honest sentence is the one the X-Factor review settled on: no scalable book survives capacity-aware costs; one low-turnover corner does, with a stated ceiling. Publishing the size at which a strategy stops working is the differentiator.

5. Where it lives (code· routes· DB· timers)

• Module: the lowvol sleeve q research code (--build writes lowvolq_book + lowvolq_holdings to the research database; --fence runs the capacity/cost recut). Monthly predecessor: the lowvol sleeve code (seal, Study 2026-07-22).
• Routes: this reference page, and a row in the Strategy library. 🔴 Deliberately no board and no published roster — see §9.
• Forward test: the ema crossover forward code carries it as the reference comparator; that runner is box-armed and fires quarterly from ~2026-10-03.

6. Data & provenance

Runs entirely on primary-source data already ingested nightly: NSE bhavcopy OHLCV (EQ+BE+BZ), corporate-action adjustments, per-symbol traded value for the liquidity and capacity work, and Nifty 500 closes for the benchmark. No vendor dependency (Guardrail #8). Volatility is computed from the adjusted price series, so it inherits the same PIT fences as every other book here.

7. Terminology canon

Sibling pages: the momentum riskadj notes (STEADY-25 / LOWVOL_MOM live there)· the union notes and the union ladder notes (the momentum family, whose variants are not fundable and say so).

8. Decision & session history

• Study 2026-07-22 — the monthly low-vol sleeve. Frictionless edge better, but its own capacity fence exposed 199%/yr turnover as the cost driver; ceiling ~₹50–100cr.
• Study 2026-07-22 — the quarterly + hysteresis v2 (registered 2026-07-23). The turnover cut bought ~5× capacity and flat drawdown. The fundable verdict.
• **** — every ratio relabelled return/vol; no number moved.
• **** — category discipline: low-vol is the excluded comparator, never a crossover result.
• X-Factor review 2026-08-05 — corrected the over-broad claim that nothing beats buy-and-hold, and found the registry defect recorded in §9.
• / 2026-08-27 — this page created; the sleeve given its first website presence.

9. Open items

1. 🔴 The forward/paper test is the gate, and it has not been judged. The ledger's own NEXT is freeze the spec, watch quarterly, before any capital. The runner fires from ~2026-10-03. Until it reports, this is a fundability verdict on in-sample-with-honest-costs evidence. 2. 🔴 The estate's own testing surface under-reports this result. research.db.strategy_runs — the table /dash/testing renders — holds one participation-cost row, the C-BLEND failure at 0.17, and zero rows for the low-vol result that passes (X-Factor review C-3). The honesty machine is mis-calibrated in the self-deprecating direction, which is as much a data defect as the optimistic kind and considerably harder to notice. 3. No roster is published. lowvolq_holdings exists only in the research database, which the web application does not read. Publishing a live roster is a SURFACE-PLAYBOOK landing in its own right — and one that should follow the forward verdict, not precede it, because a published holdings list of a fundable-looking book is the single most actionable-looking artefact this estate could put on a screen. 4. The forward runner's timer is UNCAPTURED in the repo — box-armed as hermes-ema-forward.timer while only patearn-ema-forward.service is captured under scripts/systemd/vps-live/.

10. Sources of truth

Ledger Studies 2026-07-22 / 2026-07-22 (the strategy ledger)· seals /· the module named in §5· the relabel and category ruling in the project's running record § Decision log.