Research
Signal Graveyard
Every detector ships with an evidence profile: what the anchor paper showed, where decay risk sits now, and when liquidity or microcap leakage should make the product more skeptical than the headline edge.
48 quarters41/48 quarters positive; century replication says PEAD persists but weaker
Decay mediumMicrocap-containedCapacity tight41/48 quarters positive; century replication says PEAD persists but weaker
Research Anchor
Bernard & Thomas (1989) — Post-Earnings-Announcement Drift
Akbas, Ay, Koch (2023) — PEAD Over the Past Century
Period 1974 to 1985
Capacity Posture
Event alpha is real, but capacity narrows quickly outside liquid post-earnings names.
Microcap Exposure
PEAD survives in liquid equities, but smaller names still exaggerate the historical effect.
Why It Fails
The edge decays when surprise, guidance, and revisions disagree or the post-event move gets crowded fast.
Insider Cluster
Insider Cluster IQ
457,539 insider purchases3.8% abnormal return over 21d for cluster buys vs 2.0% for non-cluster
Decay mediumMicrocap-containedCapacity tight3.8% abnormal return over 21d for cluster buys vs 2.0% for non-cluster
Research Anchor
Kang, Kim, Wang (2024) — Cluster Trading of Corporate Insiders
Cohen, Malloy, Pomorski (2012)
Period 1986 to 2016
Capacity Posture
Useful in liquid large caps, but cluster breadth is sparse and hard to scale across many symbols at once.
Microcap Exposure
Open-market insider signals are strongest where filings are informative, not where microcap illiquidity manufactures the move.
Why It Fails
Routine trades, 10b5-1 mechanics, or grant-related footnotes can make an apparent cluster informationally empty.
Peak Prox
PeakProx Momentum
90-year sampleNear-high effect explains much of momentum; crash-aware variant lifts Sharpe about 50%
Decay lowMicrocap-containedCapacity openNear-high effect explains much of momentum; crash-aware variant lifts Sharpe about 50%
Research Anchor
George & Hwang (2004) — The 52-Week High and Momentum Investing
Byun & Jeon (2023) — Momentum Crashes and the 52-Week High
Period 1926 to 2015
Capacity Posture
This is among the most scalable features because it lives in liquid price data rather than sparse corporate events.
Microcap Exposure
Momentum can look better in small names, but the 52-week-high anchor is robust enough to keep liquid implementation credible.
Why It Fails
Momentum gets hit hardest in violent rebound regimes where near-high leaders mean-revert instead of extending.
21-year sample+7.5% annual inside cheap stocks; gross profitability adds about 0.62%/month
Decay mediumMicrocap-containedCapacity open+7.5% annual inside cheap stocks; gross profitability adds about 0.62%/month
Research Anchor
Piotroski (2000) — Historical Financial Statement Information
Novy-Marx (2013) — The Gross Profitability Premium
Period 1976 to 1996
Capacity Posture
Fundamental composites scale better than many anomaly-zoo ideas, though the cleanest cheap-quality baskets can still crowd.
Microcap Exposure
Accrual and junk-value legs are more fragile in smaller names, so the composite should suppress microcap contamination explicitly.
Why It Fails
Cheap stocks stay cheap when the value leg overwhelms profitability and balance-sheet quality in a cyclical drawdown.
Gamma Fragility
Gamma Fragility
Fresh literature; sample size not stated in the seed docsLow-vs-high net gamma portfolios show about a 10% annualized spread
Decay unknownMicrocap-containedCapacity tightLow-vs-high net gamma portfolios show about a 10% annualized spread
Research Anchor
de Vries, Loeffler, Schaub (2023) — Option Gamma and Stock Returns
Barbon & Buraschi (2020) — Gamma Fragility
Period 2020 to 2023
Capacity Posture
Only liquid option chains with healthy open interest should be ranked aggressively.
Microcap Exposure
Microcap exposure is less the issue than options-market thinness; sparse OI can create false precision even in mid-caps.
Why It Fails
Dealer-state estimates break when chain mapping, stale Greeks, or concentrated open interest distort the true hedge pressure.
Signal Graveyard
Signal Graveyard
452 anomalies65% fail under stricter construction; 82% fail at the tougher multiple-testing hurdle
Decay lowMicrocap-sensitiveCapacity open65% fail under stricter construction; 82% fail at the tougher multiple-testing hurdle
Research Anchor
Hou, Xue, Zhang (2020) — Replicating Anomalies
Harvey, Liu, Zhu (2016) + McLean & Pontiff (2016)
Period 1926 to 2015
Capacity Posture
This is a meta-layer, so capacity is not the issue; the warning is about detector honesty, not trade size.
Microcap Exposure
Its core job is to expose microcap leakage before a detector is sold as scalable alpha.
Why It Fails
The graveyard itself fails if skepticism turns into a static checklist instead of consuming live calibration and liquidity evidence.
Calibration Lab
Calibration Lab
Infrastructure paper; fixed-sample coverage is the anchorConformal methods target finite-sample marginal coverage of 1-α
Decay lowMicrocap-containedCapacity openConformal methods target finite-sample marginal coverage of 1-α
Research Anchor
Guo et al. (2017) — On Calibration of Modern Neural Networks
Brier (1950) + Vovk et al. conformal literature
Period 1950 to 2017
Capacity Posture
Calibration infrastructure scales cleanly; the constraint is sample depth by bucket, not trading capacity.
Microcap Exposure
Microcap effects matter only indirectly because thin-name outcome noise can corrupt calibration if not bucketed honestly.
Why It Fails
Calibration can look healthy in aggregate while hiding thin-bucket overconfidence and stale mappings in the tails.
58 futures contractsSignificant cross-asset trend alpha; later century evidence says the regime effect is durable
Decay lowMicrocap-containedCapacity openSignificant cross-asset trend alpha; later century evidence says the regime effect is durable
Research Anchor
Moskowitz, Ooi, Pedersen (2012) — Time Series Momentum
Hurst, Ooi, Pedersen (2017) — A Century of Trend Following
Period 1985 to 2009
Capacity Posture
Macro conditioning is broad and liquid; the failure mode is stale regime inference, not trade capacity.
Microcap Exposure
Microcaps should not dominate the tape view because the evidence lives in liquid cross-asset trend measures.
Why It Fails
Macro conditioning fails when the regime flips faster than the signal refresh cadence or cross-asset cues conflict sharply.