Hypothesis spec — Gap-surge day-one fade (exploration-first)

ID: S054 Slug: gap-surge-day-one-fade-v1 Status: registered — direction only, pre-exploration. The operational trade rule is deliberately TBD until the chart review below is done. Gates are [CALIBRATE]. Nothing freezes until the first Stage-1 run starts. Created: 2026-07-18 · restructured 2026-07-18 (exploration-first, owner decision: "I want to explore first — what did the last 100 gappers of this style do — then I'll come up with a strategy") Origin: founder observation from discretionary trading (2020–2021 low-priced gapper era); systematized here precisely because the discretionary version was emotionally untradeable.

---

Order of operations (what is registered now vs. later)

1. Registered now: the directional hypothesis, the screen that defines "this style of gapper", the rationale, the execution constraints, and the contamination firewall below.
2. Exploration (internal, next step): chart review of the ~100 most recent qualifying gap days. No trading rule exists yet; the owner formulates it from this review.
3. Rule + gates written into this spec after exploration → FROZEN when the first Stage-1 run starts.
4. Stage 1 runs on pre-exploration history only (see firewall).

One-line hypothesis (direction, registered now)

We hypothesize that an opening gap-up of ≥ +20% in a sub-$20 US-listed common stock predicts negative returns between the surge-day open and the same day's close, and possibly into day +1, in the point-in-time listed spec-cap universe (delisted tickers included), because promotion-driven overnight and pre-market buying by underinformed traders systematically overprices the open, while dilutive offerings priced into strength and structurally scarce short supply delay the correction until regular trading hours.

The direction is the hypothesis. The rule (entry timing, exit timing, stops, day-1 vs day-2 handling) comes out of exploration.

Rationale (the "because")

Three mechanisms, each independently documented in the small-cap literature and consistent with practitioner experience, point the same direction:

1. Demand is promotion-driven, not information-driven. Large overnight gaps in low-priced stocks are dominated by retail flow responding to promotions, social-media attention, and momentum-chasing — buyers with no valuation anchor and no exit discipline. This flow concentrates in the pre-market and opening auction. Once the marginal promoted buyer is exhausted, there is no natural second wave of demand during regular hours.
2. Supply arrives on the surge day itself. Issuers in this segment are chronically capital-hungry; ATM offerings, shelf takedowns, and warrant exercises are disproportionately triggered on high-volume up days, because that is when the liquidity to absorb them exists. The surge creates its own overhead supply.
3. The correction is not arbitraged away because shorting is expensive. These names are low-float and hard-to-borrow; locates cost tens to hundreds of basis points per day when available at all, and many retail brokers offer no locate mechanism. Limits to arbitrage are precisely why a fade this visible can persist — and also why the friction assumptions must be punitive.

The falsification story: if the open already prices in the fade (auction efficiency), or if continuation days offset first-day fades in the pooled sample, no rule formulated from the charts will separate from baseline and the gate fails.

Screen definition (registered now — this is what "this style of gapper" means)

• Unadjusted open ≥ 1.20 × unadjusted prior close
• Unadjusted prior close in [$1.00, $20.00]
• US-listed (NYSE/NASDAQ/AMEX) common shares and ADRs; no warrants, units, ETFs, SPAC rights; OTC excluded (no data or execution path)
• Point-in-time universe including delisted tickers — this segment is a graveyard of delistings and reverse splits, and the delisted names are where the effect should be strongest
• No volume condition in the screen: full-day volume is unknowable at the open (look-ahead); pre-market volume is the live-world proxy but is not in daily data — revisited at Stage 2

The +20% threshold and $1–$20 band define the exploration sample. If the chart review shows the interesting population lives elsewhere (say +50% gappers, or sub-$5 only), re-scoping the screen for the tested rule is legitimate pre-freeze calibration — recorded here before any Stage-1 run.

Exploration protocol (pre-freeze, internal — output lives in the exploration workspace, never public)

• Sample: the ~100 most recent qualifying signal days per the screen above, working backward from today. Tickers may repeat; tag day-in-run (first gap day vs day-2/3 continuation).
• Chart pack per event: (a) daily context, ±20 sessions around the event; (b) intraday shape where available. Tiingo IEX minute bars are acceptable for eyeballing recent liquid gappers — price shape roughly tracks the consolidated tape — but they are IEX-only prints: volume is not consolidated, pre-market is largely invisible, and none of it is backtest-grade. Eyeball yes, compute no.
• Record per event: open vs close; time-of-day of the high; shape (fade-from-open vs morning-push-then-fade vs all-day trend-up); gap filled or not; day+1 open and close; trading halts; offering/dilution news where quickly findable.
• Output: an internal exploration note with the observed taxonomy and base rates. The owner formulates the operational rule from that note; the rule and its gates are then written into this spec.
• Build note: the screen is fully specified above, so the pull-and-chart job can be built directly from this file (Claude Code task; existing Tiingo EOD stack + IEX intraday endpoint).

Contamination firewall (pre-registered NOW, before any chart is looked at)

The explored events are spent. Once the exploration sample is drawn, its window — from the earliest explored event through today — is excluded from the Stage-1 test sample. Stage 1 runs on pre-exploration history only. The exploration window may be reported descriptively alongside results but cannot count toward any gate. This is the house "exploration precedes freeze" discipline (cf. S040; holdout-touch accounting on the stat-arb pillar): a rule formulated by looking at 100 charts would trivially "pass" on those same 100 charts.

Candidate baseline rule (illustrative — NOT the registered strategy)

The naive daily-bar proxy from the first draft of this spec is kept as the null baseline: short the official opening print, cover the official closing print, equal-weight per signal day, no stops. Whatever rule exploration produces should be benchmarked against this naive baseline. It may end up being the tested rule if the charts support it — or be replaced entirely.

Data required

• Tiingo EOD OHLCV, US-listed common stocks, point-in-time incl. delisted (existing stack) — for the screen, the exploration sample, and any daily-bar Stage-1 test
• Tiingo IEX intraday — exploration eyeballing only (caveats above)
• Consolidated minute data (Polygon/FirstRate class) — pre-registered prerequisite for Stage 2, and for Stage 1 too if the formulated rule is intraday (entry/exit not expressible in daily bars). Not purchased until exploration justifies it.

Gate shape (suggested only — [CALIBRATE] after exploration; FROZEN at first Stage-1 run)

Final gate metrics follow the final rule, so only the shape is registered now. For an open→close-style rule, the expected shape:

• Magnitude gate on the raw mean signal-day return
• Direction-aware significance: Welch p < 0.05 vs matched baseline (same price band, non-gap days) and signal mean on the hypothesized side
• Hit-rate floor
• Yearly-consistency floor (effect present in a supermajority of calendar years)
• Binding net gate after a punitive fixed friction model — anchor: 75 bps slippage per side + 100 bps/day borrow-and-locate proxy = 250 bps round-trip, deliberately above the house 50–100 bps spec-cap default because of locate costs. If the effect cannot pay a 250 bps toll, it is not a strategy — it is a chart pattern.
• Minimum sample floor (anchor: ≥ 300 signal days in the pre-exploration test window)

Look-ahead-trap analysis (applies to the screen; extend when the rule lands)

Known execution constraints (recorded now, so they are not a surprise later)

The current broker (tastytrade, open API confirmed workable for automated listed-equity orders) offers no locate service for hard-to-borrow shares — and nearly every name this screen fires on will be HTB. Exploration and Stages 1–2 are data-only and unaffected. Promotion beyond Stage 2 requires an execution answer: either a locate-capable broker (CenterPoint, TradeZero, Cobra class) with measured locate costs fed back into the friction model, or a re-scoped long-side variant (a new S-NNN, not an amendment). Pre-registered so a Stage-2 pass cannot quietly morph into an execution assumption the pipeline never tested.

What I expect the charts to show (written before the review)

I expect the dominant shape to be morning-push-then-fade rather than fade-from-the-open: the classic pump chart makes one or two pushes in the first 30–90 minutes, puts in the high of day before ~11:00, and closes well off the high. I expect close < open in roughly 55–60% of events, the fades to be stronger on day-2/3 continuation gaps than on first gap days, and the ugliest fades to cluster in names with offering news. If that shape holds, the natural rule will be an intraday one (short the failed second push / first lower high) — which daily bars cannot express, meaning the honest Stage-1 choice becomes either the naive open→close baseline on daily data, or buying minute data early and running Stage 1 intraday. That trade-off is decided after the review, not now. If instead the charts show no dominant shape — gappers that trend all day as often as they fade — that is a cheap early warning that the pooled edge is weaker than the discretionary-era memory suggests.