Aggregate short-interest market-timing (SII)

The question

Does aggregate short-seller positioning — sourced free from regulatory short-interest filings — predict forward equity returns and drawdowns, a short-side information set orthogonal to price, volatility, and fundamentals?

How it was tested

The hypothesis, its exact trigger, the outcome it predicts, and the pass/fail bar were written down and cryptographically hash-locked before any data was tested, then held through a mandatory cooling-off period. The backfill ran exactly once against those frozen parameters — no re-tuning, no curve-fitting — using adversarial statistics (stationary-bootstrap confidence intervals, multiple-testing correction, purged cross-validation with an embargo, and walk-forward out-of-sample splits). The specific trigger thresholds are proprietary and omitted here.

The outcome

RAN 2026-06-25 (attended) → NOISE / DISCONFIRMED. 203 bi-monthly SII obs 2017-12-29→2026-05-29 (SI 80/80, shares-out 80/80), 5 K1 onsets. PRIMARY SPX-3mo IC +0.057 — WRONG SIGN (RRZ predicts negative) + CI straddles 0 + Holm-Sidak n.s. + fails no-subsumption-vs-VIX/credit (icLowStress +0.054) + jackknife flips on COVID (icExCovid +0.062). Secondary SPX-6mo right-signed −0.031 but weak/straddles. Aggregate short interest does NOT predict forward SPX on the free ≥2017 window — the honest prior (INFO-leaning, power-limited ~3-4 bears) was charitable. Graveyard w/ yield-curve/options-skew/vol-term/net-fed-liq/macro-nowcast. Survivorship (frozen 80-ticker current shares-out applied historically) made it OPTIMISTIC → robust disconfirm. Pure info-layer — zero trade-path. `data/backfill/aggregate_short_interest_2026-06-25.json`. Rule #10: no iterate (a longer-history or per-name SI signal = NEW pre-reg). Sibling tracks (daily short-volume PEAD overlay, EDGAR net-share-issuance 6/26) unaffected. LOCKED 2026-06-20 (operator 'ok proceed', free-data catalog #1). Memo project_aggregate_short_interest_backfill_2026_06_25.md.

The backfill ran and the hypothesis failed its pre-registered bar. It is recorded and retired — not re-tested without a fresh, separately-registered hypothesis.