Swing Trading — Pillar Landscape
swing-trading-landscape-v1 · Research study · pillar landscape · status: PUBLIC · visibility: public
This is a Research study, not a setup. Research is where data is organised and priors are formed before an idea becomes a pre-registered setup. Nothing here is a validated claim or a deployable signal — it emits priors and candidate families that flow one-way into the Lab pipeline, where the frozen gates live. Published for the same reason the cuts are published: showing how the thinking forms before anything is committed.
Feeds (one-way → Lab, all pending registration): B1 short-term reversion · A1 trend-continuation breakout · C1 cross-asset RS rotation.
Scope locked: - Two families pursued as separate setup families (not merged). - Selection logic is fully mechanical (no discretionary entry). - Cross-asset from the start (equities + sector / commodity / currency / bond / country ETFs). - This document is the landscape survey only; candidate families are sketched (§8) but not pre-registered here.
0. TL;DR
Swing trading splits into two mechanically opposite families: trend-continuation / momentum (buy strength, ride a Stage-2 uptrend for weeks) and short-term mean reversion (buy a pullback inside an uptrend, exit in 3–7 days). Both are well-documented by named practitioners, and both reduce to a liquidity-gated, trend-aware, dynamically-reconstituted basket of stocks and ETFs — which is exactly the construction this pillar targets. The ETF layer is what opens commodities, currencies, bonds and countries, and is the basis of the "less correlated, sellable" thesis.
Three things make this pillar strategically attractive for Mechaniq, and one caveat must travel with it:
- It is the only pillar without the options-data problem. It runs on daily EOD stock/ETF data (Tiingo), so it is genuinely backtestable and walk-forwardable — unlike the 0DTE, wheel and iron-fly pillars, which are forward-only because Tastytrade historical options data is unreliable.
- The mechanical core is narrow. The pre-registrable part is the screen (trend + liquidity filters, RSI thresholds), not the discretionary entry that made the famous practitioners famous. The gate must sit on the narrow mechanical part.
- Mean reversion carries a fat left tail. High win rate, occasional large loss. A win-rate or magnitude-only gate would flatter it. The tail-aware gate (worst-session loss, CVaR, drawdown budget) is non-negotiable here.
- Caveat that must be sold honestly: cross-asset momentum is itself a correlated factor. The diversification is real but smaller than the pitch implies.
A prior swing-style attempt already exists in the record: S004 (rs-leadership-rotation-v1), a relative-strength rotation, was cut at S2 on a momentum-crash round-trip (+22% / −11%). That cut is part of this landscape, not separate from it.
1. Definition & horizon for Mechaniq's purposes
"Swing trading" here means multi-day to multi-week holding, signal-driven, on liquid US-listed stocks and ETFs, EOD cadence. It sits between the intraday options pillars (0DTE) and longer position trading. Two horizon bands matter and map onto the two families:
| Band | Typical hold | Family |
|---|---|---|
| Position-lean swing | weeks to a few months | Trend-continuation (Family A) |
| Short-term swing | 3–7 trading days | Mean reversion (Family B) |
These are deliberately kept as separate families because they need different baskets, different exits, and different gate shapes (Family B's fat tail in particular).
2. Family A — Trend-continuation / momentum (buy strength)
Lineage: Jesse Livermore / Richard Wyckoff → William O'Neil (CANSLIM) → Stan Weinstein (Stage Analysis) → Mark Minervini (SEPA / VCP).
Core idea. A security is always in one of four stages — Stage 1 basing, Stage 2 advancing, Stage 3 topping, Stage 4 declining — and only Stage 2 is a buy. A name that fails the trend filter is not a short; it is simply not a candidate.
The mechanical screen (pre-registrable). Minervini's Trend Template is the cleanest operational form and is essentially a basket-membership rule: - Price above the 50-, 150- and 200-day SMA. - 200-day SMA trending up for at least ~1 month. - 50-day above both 150- and 200-day; 150-day above 200-day. - Relative strength versus the market (RS line / ratio) confirming leadership.
The discretionary layer (NOT cleanly pre-registrable). Inside a Stage-2 name, Minervini waits for a Volatility Contraction Pattern (VCP): a series of progressively smaller pullbacks on drying volume, then a breakout above a pivot on expanding volume. Whether a base is a "clean" VCP is a judgement call. For Mechaniq this is the boundary: the screen is mechanical; the VCP/pivot entry must be replaced by an explicit, frozen rule (e.g. breakout above N-day high on volume ≥ X× average) or it does not belong in a signal product.
Risk management as taught. Hard stop ~7–8% below entry, partials into strength at ~2–3R, stop to breakeven once gain is established, trail along a moving average or swing-low structure. Useful as a prior for exit design; the exit must still be frozen and tested, not adopted on authority.
Instruments traded. Liquid high-momentum growth stocks (large/mid-cap); sector ETFs used as a fast screening / context layer.
Named anchors. Minervini — two-time U.S. Investing Champion (1997, 2021), audited; intermediate-term swing trader by his own description. O'Neil — CANSLIM, fundamental growth + technical timing + market-direction gate. Weinstein — Stage Analysis, the timeless price/volume/relative-strength/breadth framework the others build on.
3. Family B — Short-term mean reversion (buy weakness in an uptrend)
Lineage: Larry Connors / Linda Raschke (Street Smarts), Connors' later High Probability ETF Trading.
Core idea. Inside a confirmed uptrend, a short-term oversold reading tends to attract capital and revert. This is a tendency, not a rule — oversold can become more oversold — so it is an edge to exploit, not a certainty. Sweet spot is a 3–7 day hold, and it works best on liquid equity ETFs.
The mechanical rules (fully pre-registrable — this family's whole appeal). Connors' published systems are exact, e.g. the R3 logic: - Close above the 200-day MA (regime filter — only revert inside an uptrend). - 2-period RSI falls three days running, first day's drop from a reading below 60. - Enter when the 2-period RSI is below 10. - Exit when the 2-period RSI recovers (e.g. >70) or price closes above the 5-day MA.
Variants (RSI-25, RSI-2 < 10 / close > 200DMA) follow the same shape. Raschke's contribution is the top-down "is today a trend day or a reversion day" pre-classification and resting-stop discipline.
The structural risk — must drive the gate. Mean-reversion systems typically post very high win rates (often 80%+) but the losing trades are large, because the deeper a name falls the "better" the entry looks right up until it doesn't. This is the classic profile that a win-rate or magnitude gate rewards and a tail-aware gate correctly punishes.
Instruments traded. Liquid equity index / sector ETFs primarily; the logic generalises to other liquid instruments.
4. Practitioner reference table
| Practitioner | School | Signature technique | Instruments | Hold | Mechanical-clean? |
|---|---|---|---|---|---|
| Stan Weinstein | Trend | Stage Analysis (4 stages, buy Stage 2) | Stocks, ETFs | weeks–months | Screen yes; entry partly |
| William O'Neil | Trend + fundamentals | CANSLIM + market-direction gate | Growth stocks | weeks–months | Screen yes; entry no |
| Mark Minervini | Trend | SEPA / Trend Template / VCP breakout | Growth stocks | weeks–months | Screen yes; VCP entry no |
| Larry Connors | Mean reversion | RSI-2 systems (R3, RSI-25) | Liquid ETFs | 3–7 days | Yes (fully) |
| Linda Raschke | Both / regime | Trend-vs-reversion day classification | Futures, ETFs | days | Partly |
The pattern is consistent: trend practitioners are systematizable up to the entry; reversion practitioners are systematizable end-to-end. That is why Family B is the lower-friction first pre-registration, even though Family A is the more famous story.
5. The cross-asset ETF layer (the diversification thesis)
ETFs give one liquid ticker per sector, commodity, currency, bond complex or country, and the same swing techniques apply. This is the layer that delivers the "less correlated, new ways to make money" pitch.
Sleeves and representative tickers (illustrative, not a final universe): - Equity sectors: XLK, XLF, XLE, XLU, XLY, XLV, XLI, XLB, XLP, XLC, XLRE. - Commodities: GLD / SLV (metals), USO / XLE / XOP (energy), DBA (agriculture), DBC (broad). - Currencies: UUP (USD index), FXE (EUR), FXY (JPY) — the sleeve that most genuinely steps outside equity correlation. - Bonds: TLT (long Treasuries) — note: on the options side TLT/GLD are MWF-expiry only; on the EOD swing side they trade like any ETF. - Countries / regions: EEM, EWZ, EWS, etc.
Techniques on the ETF layer: - Relative-strength / sector rotation. Rank sector ETFs by rate-of-change over 1/3/6-month windows, hold the top 2–3, rebalance monthly, rotate out on loss of relative strength. A cleaner variant uses the ratio (e.g. XLK/SPY) rather than raw price, which strips out market beta so a "leader" is not merely "rose with SPY." - Cross-asset momentum. Measure momentum across bond / commodity / equity ETFs jointly to capture broader cycles. Academic work reports country/industry momentum is exploitable via ETFs with an excess return on the order of ~5%/yr after costs — supportive, but to be re-derived under our own gates, not taken as given.
S004 connection. The rs-leadership rotation already tested (cut at S2) is a special case of this layer restricted to equities. Two pre-registered variants exist in lab/hypotheses/ (dispersion-conditional; Sortino-gated). The cross-asset extension is the natural superset — but inherits S004's lesson: a relative-strength signature does not survive a momentum-crash round-trip unless the path/tail is controlled.
6. Basket construction (the heart of the concept)
The basket is dynamic, not a fixed watchlist. Membership is recomputed on a cadence and must be point-in-time (PIT) to avoid survivorship / look-ahead bias.
Construction pipeline:
1. Liquidity gate. Minimum average dollar volume + price floor. For ETFs, add a bid-ask spread ceiling. Determines the tradeable set. [KEUZE] thresholds.
2. Trend gate. Stage-2 / Trend-Template style filter (Family A) or the 200-DMA regime filter (Family B). Determines membership.
3. Reconstitution cadence. Weekly or monthly. [KEUZE]. Higher cadence = more turnover/cost and whipsaw; lower = staler membership.
4. PIT discipline. Membership at date t uses only data available at t. Same monthly top-N PIT universe discipline already used on the equity side. This is a hard look-ahead trap, not an optimisation.
Open construction parameters (for later setup specs, frozen at pre-registration): lookback windows for trend/RS, ranking metric, number of holdings, weighting (equal vs RS-weighted), rebalance day, and friction assumptions (default 25–30 bps round-trip for liquid large-caps/ETFs; wider for thin sleeves).
7. Mapping to the Mechaniq pipeline
Data advantage. EOD stock/ETF data is reliable and deep. This pillar can run the full S0→S4 path including S1 quick-cut and S2 walk-forward — it is not forced forward-only like the options pillars. This makes swing the strongest pillar to demonstrate the pipeline's teeth, not the weakest.
Gate philosophy (carried over, non-negotiable): - Tail-aware gate: worst observed session/trade loss, CVaR, and max drawdown must all stay within pre-registered budgets. Especially load-bearing for Family B. - Matched baseline: every claim compared against a like-for-like baseline (e.g. random sector pick, buy-and-hold the sleeve), not against zero. - Decay tripwire (mandatory for any published-edge setup): Family B's edges (RSI-2 systems) are 15+ years in the public domain; re-tests show roughly half of Connors' ETF strategies degraded after publication. Any setup derived from a public edge ships with a pre-registered decay tripwire and is cut if realised vs expected edge breaches it. - Direction-aware significance and geometric/per-fold consistency at S2 (the H-002 and arithmetic-mean lessons apply unchanged).
Naming. Each family yields one or more S-NNN setups. Variants of a tested setup become new S-NNNs, never re-runs of a frozen one.
8. Candidate setup families (sketch only — to pre-register later)
Not pre-registered here. Listed so the survey points somewhere.
- A1 — Trend-continuation breakout. Trend-Template membership + frozen breakout rule (N-day high on volume ≥ X× avg) replacing the discretionary VCP. Operationalisation risk: high (entry definition).
- B1 — Short-term reversion in uptrend. RSI-2 below threshold AND close > 200-DMA; exit on RSI recovery / 5-DMA close. Operationalisation risk: low; decay risk: high (published).
- C1 — Cross-asset RS rotation. Rank a multi-sleeve ETF universe by ROC, hold top N, rebalance. Superset of S004. Operationalisation risk: low; correlation-caveat: high.
9. Methodology risks (the honest read)
- Published-to-death edges decay. The Family B systems are public and partly worked-off. Test them anyway, but only behind a matched baseline and a decay tripwire — otherwise the product could sell an edge that is already gone.
- "Less correlated" is partly an illusion. Cross-asset momentum is one correlated factor; the sleeves co-move in trend reversals (2008-type). The diversification is real but smaller than the pitch implies, and the transparency wedge forbids overstating it.
- Practitioner edge is discretionary and survivorship-selected. We hear the winners; the "clean VCP" is judgement, not rule. The systematizable part is narrower than the literature suggests, and the gate must sit on that narrow part.
- Dynamic basket = look-ahead minefield. PIT reconstitution is mandatory; getting it wrong manufactures phantom edge.
- Mean-reversion tail. High win rate hides large losers; the tail-aware gate is the only honest test.
10. Open scope decisions [KEUZE]
Before the first setup spec:
- First family to pre-register. B1 is lower-friction (fully mechanical, smallest entry-definition risk) and is the natural first S-NNN even though A is the more famous story. Confirm order, or run A1 and B1 in parallel.
- Universe v1 breadth. Cross-asset is locked in principle — confirm whether v1 instantiates all sleeves at once or stages them (e.g. equities + sectors first, commodities/currencies as a second registered sleeve) purely to keep the first walk-forward interpretable.
- Reconstitution cadence & liquidity thresholds. Weekly vs monthly; ADV / price / spread floors.
- Baseline definition per family. Buy-and-hold the sleeve, random pick, or equal-weight benchmark — frozen before any run.
- Friction model per sleeve. Single default vs per-sleeve (thin commodity/currency ETFs deserve wider assumptions).
11. References (books / practitioners — for traceability, no quotations)
- Stan Weinstein — Secrets for Profiting in Bull and Bear Markets (Stage Analysis).
- William J. O'Neil — How to Make Money in Stocks (CANSLIM).
- Mark Minervini — Trade Like a Stock Market Wizard; Think & Trade Like a Champion (SEPA, Trend Template, VCP).
- Laurence Connors & Linda Raschke — Street Smarts; Connors & Alvarez — High Probability ETF Trading (RSI-2 systems, R3, RSI-25).
- Jack Schwager — Stock Market Wizards / The New Market Wizards (Minervini, Raschke profiles).
- Quantpedia / Quantified Strategies — independent ETF momentum and Connors-strategy re-tests (decay evidence).
Sources surveyed via web research 2026-06-27; performance figures cited by third parties are treated as claims to be re-derived under Mechaniq gates, not as established results.