Setup: Short-squeeze pressure

ID: S022 Slug: short-squeeze-pressure-v1 Type: EVT Date added: 2026-06-08 Status: blocked (data) — promoted from lab/candidates/short-squeeze-pressure-v1.md Wall: 3 — Messy data that won't license cleanly

Blocked on: scraped FINRA short interest (bi-weekly, free but stale on arrival) + borrow-fee feed (paid or scraped from IBKR / Regsho). Partial blocker: a live-trigger paper-trade variant is possible if forward-only validation is acceptable; full historical backtest needs the borrow-fee history.

One-line setup

We expect that short-dated OTM call options on small-cap names with simultaneously high short interest (≥ 20% of float) AND elevated borrow fee (≥ 5% annualised) AND a fresh upward technical trigger, sized small (defined-risk lottery-tickets), expressed over a 5-15 trading-day horizon predict positive expected value driven by short-covering as the cohort exits crowded positions in the US small-cap universe with a liquidity floor for options tradeability over the multi-week post-trigger window, because when short interest is high and the stock is expensive to borrow, shorts are under standing pressure; a small upward catalyst can force them to cover, and their covering is itself buying — a self- feeding move; the names where this bites hardest are small and hard- to-borrow, exactly the territory big funds can't operate in.

Rationale (the "because", expanded)

When short interest is high and the stock is expensive to borrow, the shorts are under standing pressure. A small upward catalyst can force them to cover, and their covering is itself buying — a self- feeding move. The names where this bites hardest are small and hard- to-borrow, which is exactly the territory big funds can't operate in.

The only long-convexity idea in the lab — it makes money in sharp up-moves, the opposite payoff shape to the short-premium book, so it hedges the book's worst environment. Critical caveat: squeezes are rare and violent; most flagged names never fire.

Data required

• FINRA short interest — free, bi-weekly publication (so stale by up to 2 weeks on average). Gate must be designed around the staleness, not pretend it's fresh.
• Borrow fee history — paid (S3 Partners, IHS Markit), free- with-effort (scrape IBKR Regsho data with daily snapshots going forward only), or accept gap (forward-only variant)
• Tiingo EOD prices + intraday for the trigger (price-momentum filter)
• Historical option chains for the OTM call backtest — vendor blocker (same as S015/S016)
• Universe filter: small-cap band, options liquidity floor

Quick-kill gate (Stage 1)

Will be considered to have passed Stage 1 if:

• Mean per-trade EV ≥ +5.0% of premium paid (high bar because hit rate is low — payoff must justify the bleed) [suggested, to freeze]
• Hit rate ≥ 20% (lottery-ticket shape; most lose, the wins must be enormous) [suggested, to freeze]
• Sample size ≥ 400 triggered events [suggested, to freeze]
• Effect present in both sample halves [suggested, to freeze]
• Trigger-validity check: results conditional on "fresh upward trigger" must beat unconditional buys on the same SI+borrow filter — confirms the technical trigger is doing real work [suggested, to freeze]
• Concentration check: no single year contributes > 40% of total P/L (catches the "one GameStop / AMC year carries everything" failure mode) [suggested, to freeze]

What I expect to find

The mechanism is real and the tail-events are dramatic, but the hit-rate-vs-premium-bleed math is brutal. Probability of clearing the gate is moderate-low (~30-40%). Most likely failure modes: (a) concentration check fires because 2020-21 GME/AMC era carries disproportionate P/L and the rest of the sample is bleed; (b) trigger-validity check shows no incremental edge — the SI+borrow filter alone may be doing all the work, with the trigger adding nothing.

Notes

• Live-trigger paper-trade variant is the alternative path if the borrow-fee history blocker can't be resolved. Run 12-16 weeks of forward paper-trade on the live SI+borrow+trigger gate; the sample drops to ~30-60 events but the data dependencies become much lighter.
• FINRA staleness: the bi-weekly publication means short interest on day D may be ~10 trading days old. Gate must use the most recent published value at decision time, not a hindsight value.
• Crowded shorts can stay crowded a long time before anything happens — bleed time-decay is the dominant cost. Small defined- risk position-sizing is non-negotiable.

Disclosure boundary

This setup file is internal. Downstream result.md / kill.md writeups must follow lab/DISCLOSURE_POLICY.md §2. Pre-publish: python -m pytest tests/test_disclosure.py.