Remove ads · Upgrade

← Methodology

87

families

16

themes

17

papers linked

69

data sources

21

recently added

Every family is grounded in a peer-reviewed paper. We backtest, validate, and continuously re-run them across 4,900+ US-listed tickers.

Alpha Family Catalogue

Every alpha we run, with the paper it came from.

87 factor families across 16 themes — each one grounded in a peer-reviewed paper or quantitative research note. Click any family for its signal rule, data dependencies, paper citation, and example tickers.

Momentum (13)Mean-Reversion (5)Pairs (2)Quality (2)Accounting (5)Event (8)Insider-Activity (2)Sentiment (7)Text-NLP (12)Macro (8)Short-Flow (5)Microstructure (3)Risk-Premium (4)Diffusion (1)Network-Effects (4)Meta (6)

Featured family of the weekMacro

`vrp vix term`

VIX term-structure slope (VIX9D / VIX vs VIX / VIX3M) is a coincident indicator of market regime. Stay long the stock when the term structure is in contango (VIX < VIX3M) — backwardation signals

#19 VRP / VIX term structure (Bollerslev, Tauchen, Zhou 2009).

Coverage by theme

Insider-Activity

Network-Effects

Papers by decade

Pre-2000

7

2000s

16

2010s

29

2020s

18

Our oldest signal is from 1973; newest from 2025. 37% of families are based on papers published since 2015.

New only

Showing 87 of 87 families

Momentum

13 families

#28atr breakout

Different from #7 breakout_volume (which keys off volume confirmation): ATR breakout uses Average True Range to size the breakout magnitude in units of recent realized vol. A 1.5×ATR break above the 20-day high is the canonical Donchian / Turtle entry — Curtis Faith's "Turtle Traders" edge on trend

Different from #7 breakout_volume (which keys off volume confirmation):

#3breakout proximity

Mechanism — see detail page.

—

#7breakout volume

Mechanism — see detail page.

—

#2cross sectional momentum

For a single ticker, the cross-sectional analog is: rank the stock vs a peer/sector basket on trailing 12m return ex-1m, go long when in the top tercile of its peer group. Implemented per-ticker by comparing to the sector ETF's universe — without explicit peer-by-peer ranks we use the stock vs sector ETF residual return as a

#2 Cross-sectional momentum — Jegadeesh-Titman 1993.

Edge: Sharpe declined to ~0.3 since 2010 — survives but crowded

#77frog in pan momentum

Da, Gurun & Warachka (2014) *Review of Financial Studies*, "Frog in the Pan: Continuous Information and Momentum."

Da, Gurun & Warachka (2014) *Review of Financial Studies*, "Frog in the

Edge: 6%/yr for continuous winners vs −2% for discrete (Da-Gurun-Warachka 2014)

#29high 52w momentum

George & Hwang (2004) and Geczy & Samonov (2015) show that the proximity to the 52-week high is a stronger predictor of next-month returns than raw 6-month momentum. Stocks within 5% of their 52-week high earn ~0.45%/month more than the

George & Hwang (2004) and Geczy & Samonov (2015) show that the

Edge: Stocks within 5% of their 52-week high earn ~0.45%/month more than the market.

#12idiosyncratic momentum

Strip out market+factor beta from stock returns via OLS on Fama-French factors, then rank by trailing residual return. Position long when residual is

#12 Idiosyncratic / Residual Momentum (Blitz, Hanauer, Vidojevic 2020).

#11jolts hiring momentum

Belo, Lin & Bazdresch (2014) *J. Acc. Econ.* — "Labor hiring, investment, and stock return predictability". Firms (and industries) with strong hiring growth tend to outperform over the following 1-2 quarters because hiring growth precedes earnings growth (management is acting on private information about

Belo, Lin & Bazdresch (2014) *J. Acc. Econ.* — "Labor hiring, investment,

Edge: • LONG when industry openings YoY > +20% (strong sectoral demand).

#1multi horizon trend

Mechanism — see detail page.

—

#17sector momentum

Often stronger than stock-level momentum: long the stock when its sector ETF is in a strong

#17 Industry/sector momentum — Moskowitz-Grinblatt 1999.

#68tnic momentum spillover

Lee, Sun, Wang & Zhang (2023) *RFS*: "Technology Spillovers and Cross- Predictability." Lagged 1-month return of a focal stock's TNIC-3 peer basket predicts the focal stock's next-month return. Sharpe 1.3, +9% annualised alpha after Fama-French 5. The signal is orthogonal to industry/sector momentum because TNIC peers cross industry boundaries (text-similarity not…

Lee, Sun, Wang & Zhang (2023) *RFS*: "Technology Spillovers and Cross-

Edge: Sharpe 1.3, +9% annualised alpha after Fama-French 5.

Sign of trailing 12-month return × inverse-volatility scaling. The canonical hedge-fund trend signal — different from "multi-horizon trend voting" because it explicitly *vol-scales*

#11 TSMOM — Time-Series Momentum (Moskowitz, Ooi, Pedersen 2012).

Edge: 0.7-1.0 Sharpe across asset classes 1985-2009; ~0.3 single-stock OOS

#4vol timed ma

Per the literature, technical rules behave differently across vol regimes; gating a moving-average crossover on the realized-vol bucket makes the rule fire only when conditions historically favor

Per the literature, technical rules behave differently across vol regimes;

Mean-Reversion

5 families

#48crowding reversal

liveMean-Reversion

Stocks that mutual funds + 13F filers are forced to buy/sell because of their own client flows (not because of fundamentals) experience reversal over the subsequent 1-3 months. Heavily-owned-by-active-funds names underperform after extreme inflows; lightly-owned names outperform after extreme

#48 crowding_reversal — Wardlaw 2020 fund-flow induced trading.

liveMean-Reversion

Two well-documented intraday/swing

Two well-documented intraday/swing edges:

Edge: Most informative when the gap follows a consolidation breakout (price was within 5% of 20-day high).

#15lottery max

liveMean-Reversion

High recent maximum daily returns predict underperformance — used as a fade signal: stay flat (or short) when max-1m daily return is in the top

#15 MAX / lottery filter — Bali, Cakici, Whitelaw 2011.

Edge: ~1% per month long-short (Bali-Cakici-Whitelaw 2011)

#6range regime meanrev

liveMean-Reversion

Mechanism — see detail page.

—

#18short term reversal

liveMean-Reversion

Mechanism — see detail page.

#18 Short-term reversal — sub-week mean reversion (Jegadeesh 1990).

Pairs

2 families

#20pairs cointegration

Tests stock vs sector ETF for cointegration before trading the spread. Without cointegration the z-score spread is

#20 Pairs cointegration — Engle-Granger 2-step (Gatev, Goetzmann, Rouwenhorst 2006).

#5pairs relative value

Light-touch single-leg implementation: when the stock z-spread vs its sector ETF deviates beyond ±2σ, fade. Without explicit pair selection this is a simpler "relative value vs benchmark"

Light-touch single-leg implementation: when the stock z-spread vs its

Quality

2 families

For a single stock, "BAB strategy" can be expressed as: scale exposure inversely to rolling beta vs SPY. Lower-beta moments → bigger position; high-beta moments → reduced. The stock-level analog of the BAB

#16 BAB — Betting Against Beta (Frazzini & Pedersen 2014).

Edge: ~0.7 Sharpe in Frazzini-Pedersen 2014; ~0.4 in recent OOS

Combine quality (gross profit / assets, ROE) with momentum: long when both are in their respective top buckets relative to their own history. Stock-level adaptation; the cross-sectional implementation would rank across the universe but we approximate with self-relative

#21 Quality-Momentum (Asness, Frazzini, Pedersen 2019 — QMJ).

Edge: 0.6-0.9 Sharpe globally (Asness-Frazzini-Pedersen 2019)

Accounting

5 families

#36buyback drift

Companies announcing open-market share repurchases earn ~3-4% abnormal return over the subsequent 12 months. Effect is strongest in "value-distressed" announcers (book-to-market in the top quartile, recent price

#36 buyback_drift — Ikenberry-Lakonishok-Vermaelen 1995 / Peyer-Vermaelen 2009.

Edge: Companies announcing open-market share repurchases earn ~3-4% abnormal return over the subsequent 12 months.

#33cash operating profitability

Distinct from QMJ's "profitability" leg: cash-based operating profit (operating income + depreciation − ΔAR − ΔInventory + ΔAP) / book equity is a stronger single-name predictor than Novy-Marx's gross profitability or GAAP operating

#33 cash_operating_profitability — Ball-Gerakos-Linnainmaa-Nikolaev 2016.

#38novy marx gross profitability

(Revenue − COGS) / total_assets — gross profitability — predicts cross-sectional returns as strongly as book-to-market, and has been the mainstream "profitability" leg of the FF5 model since 2015. Distinct from QMJ which composites profitability+growth+safety; standalone GP is the high-beta version that explains more of the

#38 novy_marx_gross_profitability — Novy-Marx 2013.

Edge: Decay debate is whether the spread has narrowed (yes: 30-50%) but the sign is intact and replicable across vendors.

#39rd capitalized value

Standard book-to-market based value factors mis-classify R&D-heavy firms as "growth" because R&D is expensed (not capitalized) under GAAP. Peters-Taylor 2017 + Eisfeldt-Kim-Papanikolaou 2020 show that adding capitalized R&D back to book equity restores the value premium for tech/biotech and reverses the apparent "value-is-dead" finding of the

#39 rd_capitalized_value — Peters-Taylor 2017 intangible-adjusted value.

#37sloan accruals

Earnings persistence is much lower in firms with high accruals. Going long low-accrual and short high-accrual delivers ~10% annualized over 1962-1991 (Sloan 1996, Accounting Review). Richardson-Sloan-Soliman- Tuna (2005, JAE) refined the measure to a broader balance-sheet

#37 sloan_accruals — Sloan 1996 accruals anomaly.

Edge: ~10% ann. long-short in Sloan 1996; 4-6% post-publication

Event

8 families

#32earnings announcement premium

Stocks earn an abnormally positive return in the days surrounding scheduled earnings announcements (T−2 to T+1). The original 2007 paper documented ~9bp/day average premium during the window vs the non-announcement baseline; later replications show the effect has compressed to ~5-6bp/day but is still

#32 earnings_announcement_premium — Frazzini-Lamont 2007 EAP.

#9event aware

This generator pairs a base trend signal with an earnings-blackout filter: no entries within ±N bars of an earnings announcement, where IV/realized mismatch typically washes out

This generator pairs a base trend signal with an earnings-blackout filter:

#52index rebalance drift

When a stock is added to the S&P 500, index funds must buy it on the effective date. The announcement (typically 3-5 trading days before) triggers a front-running rally averaging +8% by effective date. Deletions show a symmetric -4% drop. Both effects partially reverse in the 20 days after effective

Crowd*, Financial Analysts Journal 2023 (avg 8.08% add-day pop, 4.85%

Edge: Ref: Chen, Singal, *Earning Alpha by Avoiding the Index Rebalancing Crowd*, Financial Analysts Journal 2023 (avg 8.08% add-day pop, 4.85% reversal).

#46lazy prices

Stocks whose 10-K (or 10-Q) text barely changes year-over-year OUTPERFORM those with big language shifts. The intuition: boring filings ≈ stable business ≈ slow-and-steady cash flow. Big text changes signal management hiding bad news with new

#46 lazy_prices — Cohen-Malloy-Nguyen 2020.

Edge: ~22% ann. long-short in Cohen-Malloy-Nguyen 2020; ~0.4 Sharpe in 2020-2024 OOS

lazy prices short section

Firms rewriting 10-K sections year-on-year signal regime change. Cohen et al. find shorting 'changers' yields up to 188bps/month — the asymmetry is strong: most alpha lives in the SHORT leg. Adds section-resolved diff (Item 1A risk-factors, Item 7 MD&A) over whole-document cosine.

Cohen, Malloy & Nguyen 2020, Journal of Finance, 'Lazy Prices'

Edge: 22% ann. long-short in Cohen et al.; doubles current long-only edge

#47m and a arb

When a ticker files an 8-K with Item 1.01 (Entry into Material Definitive Agreement) or Item 2.01 (Completion of Acquisition or Disposition), the stock often experiences directional drift over 1-3 months as the deal plays out or the market

When a ticker files an 8-K with Item 1.01 (Entry into Material Definitive

Buy after earnings surprise > 1σ, hold 30-60 days. Surprise = (actual − consensus) /

#13 PEAD — Post-Earnings Announcement Drift (Bernard-Thomas 1989).

Edge: ~9% ann. on top decile SUE (Bernard-Thomas 1989); robust 4-5% in modern OOS

strategic friday pead

Managers schedule bad-news earnings for Fridays after-close. Three-day abnormal returns around the scheduling announcement are significantly negative. PEAD persists up to 2 years for Friday-bad-news.

deHaan, Shevlin & Thornock 2015, J. Acc. Econ.

Edge: Modifier ~0.4 Sharpe boost on PEAD short side

Insider-Activity

2 families

#31insider form4

liveInsider-Activity

Lakonishok-Lee 2001, Cohen-Malloy-Pomorski 2012: corroborated insider **purchases** (Form 4 transaction_code='P') from multiple insiders within a short window are the single most-predictive insider signal — ~6%/year alpha over 6-12 months. Single-insider buys are noisy; cluster buys (3+ distinct insiders, or a single buy > $1M from CEO/CFO) are

Lakonishok-Lee 2001, Cohen-Malloy-Pomorski 2012: corroborated insider

Edge: ~6%/yr alpha for cluster buys; 82bps/mo when filtered for opportunistic insiders

insider routine filter

pilotInsider-Activity

Each insider has a 'routine' footprint (monthly/quarterly 10b5-1 clockwork buys) vs 'opportunistic' trades. Filter OUT routine insiders — the remaining cluster buys predict 82bps/month. Routine alone is zero.

Cohen, Malloy & Pomorski 2012, Journal of Finance, 'Decoding Inside Information'

Edge: 82bps/month VW alpha on opportunistic-filtered (doubles signal-to-noise)

Sentiment

7 families

analyst dispersion

High analyst-forecast dispersion → asymmetric overpricing (short-sale constraints keep pessimists out). Short high-dispersion / long low-dispersion → 7-8% annualized.

Diether, Malloy & Scherbina 2002, Journal of Finance

Edge: ~7% ann. historically; mechanism robust to short-sale constraint regimes

#78analyst forecast dispersion

Diether, Malloy & Scherbina (2002) *Journal of Finance*, "Differences of Opinion and the Cross-Section of Stock Returns."

#78 analyst_forecast_dispersion — Diether-Malloy-Scherbina 2002.

Edge: longing bottom-decile delivered ~7-9% annualized — survives Hou-Xue-Zhang (2020) replication because the short-constraint mechanism is structural, not stale arb.

#66analyst revision jump

So & Wang (2023) *Journal of Accounting Research*: "News-Implied Analyst Revisions and Drift." A large overnight gap on day t that ISN'T preceded by an analyst revision is mispriced — the revision arrives in T+5 and the price continues to drift through the revision. Reported: Sharpe 1.5, 30-day drift, robust

So & Wang (2023) *Journal of Accounting Research*: "News-Implied Analyst

Edge: Flag gaps where |gap| >= 3% AND no earnings event in trailing 5d.

#45attention spike

Abnormal Google search-volume on a ticker (SVI z-score) predicts: • SHORT-RUN OVERPRICING: investor attention spike → retail piles in → ~2-4 week mean-reversion as price corrects. • Especially in retail-heavy names (small/mid-cap, meme

#45 attention_spike — Da-Engelberg-Gao 2011 "In Search of Attention".

Edge: 2-week long for spike, 6-month short basket for unwinding (Da-Engelberg-Gao 2011)

wayback homepage diff

exploratorySentiment

Marketing/product-launch URLs appearing on the homepage precede product-revenue inflection. New CEO/CFO bios appearing predict regime change.

Implied from RenTec / Two Sigma alt-data culture (DE Shaw, Eagle Alpha, Crimson Hexagon variants)

Edge: Anecdotal — pilot required before sizing. Lower confidence.

wikipedia attention

Abnormal retail attention → 2-week price run-up, year-out reversal. Wikipedia pageviews is a cleaner orthogonal signal than Google Trends (Pyun 2024 shows it dominates Trends in OOS tests).

Da, Engelberg & Gao 2011, J. Finance; Pyun 2024 (SSRN 5172055)

Edge: 2-week long ~3%/month; 6-month short basket on attention-without-earnings

#10wikipedia attention spike

Da, Engelberg, Gao (2011, J. Finance) "In Search of Attention" established that abnormal retail attention predicts a 2-week price run-up followed by a year-out reversal. Pyun (2024, SSRN 5172055) extends the work to Wikipedia pageviews and shows the Wiki signal dominates Google Trends in OOS tests — no rate limit, no normalisation-window confound, cleaner page identity per…

#10 wikipedia_attention_spike — Da-Engelberg-Gao 2011 + Pyun 2024.

Text-NLP

12 families

#59transcript ai exposure

Eisfeldt, Schubert & Zhang (2023) NBER w31222: "Generative AI and Firm Values". HHVT-style bigram score against an AI/LLM training corpus (ChatGPT-era ML papers + tech-press articles). Documented a 5-month L/S of ~9% post Nov-2022 on the AI-exposure decile — the cleanest paper-published AI-narrative trade in the

Eisfeldt, Schubert & Zhang (2023) NBER w31222: "Generative AI and Firm

Edge: Documented a 5-month L/S of ~9% post Nov-2022 on the AI-exposure decile — the cleanest paper-published AI-narrative trade in the literature.

#58transcript climate risk

Sautner, van Lent, Vilkov & Zhang (2023) *Journal of Finance* 78(3): "Firm-Level Climate Change Exposure". Same HHVT-style bigram methodology, applied to climate-specific training corpora (IPCC reports + ESG filings). The published lexicon distinguishes three sub-exposures: physical (extreme weather, sea-level rise), regulatory (carbon pricing, emissions caps), and opportunity…

Sautner, van Lent, Vilkov & Zhang (2023) *Journal of Finance* 78(3):

Edge: Long-short on the opportunity decile earned ~6% annualised post-2015 in the paper's sample.

#56transcript exec uncertainty

Loughran-McDonald 2011, Demers et al. 2021 (Frontiers in AI): the uncertainty/weak-modal word rate within EXECUTIVE turns (not the full transcript) is a sharper bearish signal than overall transcript sentiment. The CEO/CFO speech is filtered by speaker regex (chief executive, chief financial, ceo, cfo, president,

Loughran-McDonald 2011, Demers et al. 2021 (Frontiers in AI): the

#61transcript finbert qa dispersion

Huang, Wang & Yang (2023) *Contemporary Accounting Research*: "FinBERT: A Large Language Model for Extracting Information from Financial Text." FinBERT-tone outperforms Loughran-McDonald wordcount on every post-call drift measure they tested (3-day CAR R² roughly doubles). Follow- up by Chen et al. (2024 *JFE*) shows a ~5% annualised long-short on the Q&A-dispersion

Huang, Wang & Yang (2023) *Contemporary Accounting Research*:

Edge: (2024 *JFE*) shows a ~5% annualised long-short on the Q&A-dispersion sub-signal.

#55transcript guidance shift

Bowen-Davis-Matsumoto 2002, Demers-Vega 2010: keyword markers of guidance direction predict drift. We count occurrences of direction-specific phrases per call and z-score within own

Bowen-Davis-Matsumoto 2002, Demers-Vega 2010: keyword markers of

#60transcript obfuscation bloomfield

Bloomfield (2002) Incomplete Revelation Hypothesis: managers obfuscate bad news through reduced readability. Bochkay-Chychyla-Nanda (2020) *JAR* formalized the test on conference calls: the Gunning Fog Index Q/Q delta predicts 12-month negative drift on the readability-decay decile (~4% underperformance

Bloomfield (2002) Incomplete Revelation Hypothesis: managers obfuscate

Edge: Bochkay-Chychyla-Nanda (2020) *JAR* formalized the test on conference calls: the Gunning Fog Index Q/Q delta predicts 12-month negative drift on the readability-decay decile (~4% underperformance annualised).

#62transcript peer relative tone

Frankel, Jennings & Lee (2022) *RAST*: "Disclosure Sentiment: Machine Learning vs Dictionary Methods" — FinBERT-tone residualized against peers in the same industry has 2× the post-earnings drift R² of raw tone. The idea: an absolute tone of +0.2 means very different things for a defensive consumer-staples name (peer baseline ~+0.3) vs an early-stage biotech (peer baseline…

Frankel, Jennings & Lee (2022) *RAST*: "Disclosure Sentiment: Machine

#57transcript political risk hhvt

Hassan, Hollander, van Lent & Tahoun (2019) *Quarterly Journal of Economics* 134(4): "Firm-Level Political Risk: Measurement and Effects". The authors build a political-bigram lexicon from training corpora (politics textbooks + party platforms) and score each conference call as the share of its bigrams that appear in the lexicon. The score predicts hedging activity, capex…

Hassan, Hollander, van Lent & Tahoun (2019) *Quarterly Journal of Economics*

Edge: The score predicts hedging activity, capex sensitivity to elections, and — in the 2023/2024 updates (Hassan-Schreger- Schwedeler-Tahoun JF 2023) — cross-sectional returns: long-short decile spread ≈ 4–6% annualised, Sharpe 0.6–0.8.

#54transcript qa combativeness

Hollander-Pronk-Roelofsen 2010, Matsumoto-Pronk-Roelofsen 2011: longer analyst questions + lower management answer/question word-count ratios signal skepticism — and predict near-term underperformance. We compute two per-call

Hollander-Pronk-Roelofsen 2010, Matsumoto-Pronk-Roelofsen 2011: longer

#65transcript qa evasion

Bochkay, Brown, Leone & Tucker (2024, forthcoming *JAR*): "Managers' Use of Language in Earnings Conference Calls and Future Performance." The signal is the semantic distance between an analyst's question and management's answer. When the answer is far from the question (mgmt evading), the firm underperforms over the next 60

Bochkay, Brown, Leone & Tucker (2024, forthcoming *JAR*): "Managers' Use of

Edge: Reported: LS quintile Sharpe 1.4, 7.8% annualised, 60-day horizon.

transcript text features

We use a small curated subset of Loughran-McDonald (2011) finance dict plus a few standard modal markers. Full LM dict has ~4k words; this ~300-word subset captures the high-signal common cases without bloating the package. If we want production-grade LM scoring, drop a CSV of the full dictionary in `services/worker/data/loughran_mcdonald.csv` and swap `_load_lm_terms()` to…

We use a small curated subset of Loughran-McDonald (2011) finance dict

#53transcript tone drift

Loughran-McDonald 2011, Price et al. 2012: changes in management's linguistic tone on earnings calls predict near-term price drift. We compute a per-call LM tone = (positive - negative) / (positive + negative) on the FULL transcript, then signal on the quarter-over-quarter

Loughran-McDonald 2011, Price et al. 2012: changes in management's

Macro

8 families

#30calendar anomalies

Three robust calendar effects from academic

• TURN-OF-MONTH (Ariel 1987, Lakonishok-Smidt 1988):

Edge: • PRE-FOMC DRIFT (Lucca-Moench 2015): the 24h before scheduled FOMC announcements show abnormally high returns (~3-5% annualized excess).

hiring momentum

Hiring growth precedes earnings growth by 1-2 quarters. Firms aggressively hiring signal management's private info about demand.

Belo, Lin & Bazdresch 2014, J. Acc. Econ.

Edge: 4-6% ann. at firm level; weaker at industry level

lobbying shock

Lobbying is a leading indicator of regulatory tailwinds the firm expects to win — pharma facing FDA, finance facing CFPB. Changes in quarterly lobbying spend predict 6-12 months ahead returns.

Chen, Parsley & Yang 2010 + follow-ups

Edge: 4-6% long-short historically; best in mid-caps

#44macro regime

Beyond `regime_overlay` (#10) which uses VIX + SPY trend, this family keys off the *macro* regime detected from FRED

Beyond `regime_overlay` (#10) which uses VIX + SPY trend, this family

#50mag7 factor overlay

AQR's 2025 "A New Paradigm in Active Equity" identifies Mag-7 concentration as its own factor: when the Mag-7 (AAPL/MSFT/NVDA/GOOGL/ META/AMZN/TSLA) outperform in concert, the rest of the market lags mechanically (index-weight rebalancing flow), and vice

AQR's 2025 "A New Paradigm in Active Equity" identifies Mag-7

Edge: Signal construction (own-ticker variant — not the Mag-7 themselves): • Compute Mag-7 equal-weight return series from daily_prices • Rolling 60d Mag-7 cumulative return → mag7_trend • Variants: (a) "rotation_away": LONG ticker when Mag-7 trend > +5% AND ticker is NOT Mag-7 AND t

#51pre fomc drift

Pre-FOMC drift: SPX abnormally returns ~+0.50% in the 24h before each scheduled FOMC announcement (1994-2011 sample, Lucca-Moench 2015 NY Fed SR-512). The 2024 Applied Economics update finds the drift persists post-2011 but concentrates on press-conference meetings (which is every meeting since

#51 pre_fomc_drift — Lucca-Moench 2015 + 2024 update event-driven version.

Edge: ~3.9% ann. from this drift alone (Lucca-Moench 2015)

#10regime overlay

Gate exposure to a stock by SPY trend + VIX level. Stay long only when SPY is above 200d MA AND VIX is below its rolling

Gate exposure to a stock by SPY trend + VIX level. Stay long only when

#19vrp vix term

VIX term-structure slope (VIX9D / VIX vs VIX / VIX3M) is a coincident indicator of market regime. Stay long the stock when the term structure is in contango (VIX < VIX3M) — backwardation signals

#19 VRP / VIX term structure (Bollerslev, Tauchen, Zhou 2009).

Edge: ~6%/yr unconditional, 1.0 Sharpe conditional (Bollerslev-Tauchen-Zhou 2009)

Short-Flow

5 families

#63borrow ftd squeeze

Beschwitz, Honkanen & Schmidt (2024) *Journal of Financial Economics*: "Costly Arbitrage and the Short-Squeeze Premium." Novel finding: when all three short-side stress signals fire simultaneously — top-decile borrow-rate Δ, Reg-SHO threshold inclusion, and elevated FINRA short-volume ratio — the next 5 trading days mean-revert sharply higher. The 3-table confluence is what…

Beschwitz, Honkanen & Schmidt (2024) *Journal of Financial Economics*:

Edge: Reported: 18% annualised, Sharpe 1.6 post-cost, 5-day hold.

#25borrow rate spike

When the cost to borrow a stock spikes, it

When the cost to borrow a stock spikes, it signals:

Edge: ~10% ann. on shorting expensive-to-borrow (Engelberg-Reed-Ringgenberg 2018)

#26ftd threshold list

The SEC's Threshold Security List names securities that have had 5+ consecutive settlement days with failures-to-deliver (FTDs) ≥ 10,000 shares AND ≥ 0.5% of shares outstanding. Inclusion on the

The SEC's Threshold Security List names securities that have had 5+

Edge: The SEC's Threshold Security List names securities that have had 5+ consecutive settlement days with failures-to-deliver (FTDs) ≥ 10,000 shares AND ≥ 0.5% of shares outstanding.

#24short interest change

Two complementary readings of FINRA's daily short-sale

Literature: Boehmer-Huang-Jiang 2010 ("Short Sellers Are Informed"),

#67transcript tone x short interest

Cookson, Engelberg & Mullins (2023) *RFS*: "Echo Chambers" + Jiang-Li (2024) extension. The compound signal: a negative tone shift on the earnings call followed by rising FINRA short-volume in the next 5 days predicts continued downside over a 20-day hold. The cross-table version adds ~30% to plain transcript-tone alpha because short-flow corroborates the negative narrative —…

Cookson, Engelberg & Mullins (2023) *RFS*: "Echo Chambers" + Jiang-Li

Edge: The cross-table version adds ~30% to plain transcript-tone alpha because short-flow corroborates the negative narrative — the market is acting on it, not just reading it.

Microstructure

3 families

#8flow confirmed

liveMicrostructure

Uses FINRA short-volume + ATS dark-pool data we backfilled (free) to gate a base trend signal. Long only when short_pct < median AND dark-pool volume is rising — both indicate institutional

Uses FINRA short-volume + ATS dark-pool data we backfilled (free) to gate

liveMicrostructure

Per-ticker historical IV at strike level requires a paid vendor (Polygon Options Starter ~$29/mo or OptionsDX bulk one-shot). For now this generator emits a single failed candidate per ticker so the experiment table records what's

#14 IV skew / put-call skew (Xing, Zhang, Zhao 2010) — vendor-blocked skeleton.

#22ofi microstructure

liveMicrostructure

True OFI requires L1 NBBO quote ticks. Polygon Stocks Advanced ($199/mo) or Databento pay-per-query are the path. Skeleton emits a failed candidate per ticker so the data gap is captured in the experiment

#22 Microstructure / OFI (Cont, Kukanov, Stoikov 2014) — vendor-blocked skeleton.

Risk-Premium

4 families

#34idio vol puzzle

liveRisk-Premium

Stocks with high idiosyncratic volatility relative to FF3 earn LOWER returns going forward — the "IVOL puzzle". Counterintuitive (more idio risk should imply higher required return) but robust across 23+

#34 idio_vol_puzzle — Ang-Hodrick-Xing-Zhang 2006/2009.

Edge: Two complementary readings: • SHORT HIGH-IVOL — top-quintile IVOL gives ~−7% annualized excess return after FF3 controls.

#40max drawdown premium

liveRisk-Premium

Stocks with the worst trailing left-tail outcomes (deepest 1y max drawdown, fattest CVaR(5%)) earn HIGHER subsequent returns — risk compensation for crash exposure. Standard finance ("low-risk anomaly") would predict the opposite; this is one of the few risk factors that actually points the textbook

#40 max_drawdown_premium — Atilgan-Bali-Demirtas-Gunaydin 2020.

Edge: Status: contested in pure form (overlaps with distress / O-score) — once you control for default-risk proxies, the residual sleeve is ~2-3% annualized but stable.

#49realized skew xs

liveRisk-Premium

Stocks with negative realized skewness (left-tail risk, fat downside) trade at a discount and offer a return premium going forward. Conversely positive-skew "lottery" stocks (right-tail upside) are overpriced and

Stocks with negative realized skewness (left-tail risk, fat downside)

#41speculative beta

liveRisk-Premium

Distinct from BAB (#16): Hong-Sraer 2016 (JF) show that the underperformance of high-beta stocks is concentrated in periods of high disagreement (proxied by analyst forecast dispersion OR by realized vol dispersion across estimation windows). Low-disagreement high-beta is fine; high-disagreement high-beta

#41 speculative_beta — Hong-Sraer 2016 speculative betas.

Diffusion

1 family

#35industry lead lag

Information about industry-A fundamentals diffuses slowly to firms in industry-B that are economically linked. Menzly-Ozbas (2010, JF) show upstream industry returns predict downstream industry returns at 1-month horizon, with the slowest diffusion in retail-light / illiquid

#35 industry_lead_lag — Menzly-Ozbas 2010 customer-supplier lead-lag.

Edge: Signal: • LONG when sector ETF outperformed SPY by 5%+ over trailing month AND own-ticker lagged sector by 2%+ — the laggard catches up.

Network-Effects

4 families

buffett clone 13f

pilotNetwork-Effects

A small set of 'smart' managers (Berkshire, Greenlight, Pershing, Baupost, Scion, Appaloosa) hold concentrated, high-conviction names. Their Top-1 overweight relative to benchmark outperforms their full portfolio.

Cohen, Polk & Silli 2010 + Frazzini-Kabiller-Pedersen 2018 'Buffett's Alpha'

Edge: ~4% ann. alpha (Cohen-Polk-Silli); 24% ann. in public Quantpedia replication

customer momentum supply chain

pilotNetwork-Effects

When a firm's major customer has a return shock, the supplier's stock drifts in the same direction over the following month — investors fail to update on the link. 113bps/month alpha.

Cohen & Frazzini 2008, J. Finance, 'Economic Links and Predictable Returns'

Edge: 113bps/month original; modern 50-70bps/month

patent innovation premium

pilotNetwork-Effects

Patent values inferred from 3-day stock reaction to USPTO grant. Firms with high value of recently granted patents (relative to market cap) outperform by ~3-5%/yr — real options on growth the market under-prices until citations validate.

Kogan, Papanikolaou, Seru & Stoffman 2017, QJE (extended to 2024)

Edge: 3-5% ann. long-short; survives most replication tests

#79smart money best ideas

liveNetwork-Effects

Cohen, Polk & Silli (2010), "Best Ideas." https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1364827 Frazzini, Kabiller & Pedersen (2018), "Buffett's Alpha."

Cohen, Polk & Silli (2010), "Best Ideas." https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1364827

Edge: a benchmark* — the names they actually have conviction in — produce 4-9% annual alpha.

Meta

6 families

champion overlay

Take this ticker's most recent ``user_strategies`` champion, replay its base position from the recorded indicator templates, then apply each Alpha Discovery V2 overlay filter on

Take this ticker's most recent ``user_strategies`` champion, replay its base

#64champion overlay disagreement

Bryzgalova, Pelger & Zhu (2025) *Journal of Finance*: "Forest Through the Trees" — when an ensemble of model-derived signals disagrees on direction for the same asset, the disagreement itself is informative. High disagreement flags high regime-uncertainty rows where position size should be reduced or extremes

Bryzgalova, Pelger & Zhu (2025) *Journal of Finance*: "Forest Through the

Edge: show disagreement spikes mark the inflection point of consensus reversal — fading into the disagreement earns ~20-30% Sharpe lift as an overlay on the base book).

#42meta equal weight

DeMiguel-Garlappi-Uppal (2009, RFS) "Optimal vs Naive Diversification" shows that 1/N portfolio weighting is shockingly hard to beat out-of-sample once you account for estimation error in fancy optimization. We apply this lesson at the alpha-family level: compute K canonical sub-signals from prices alone and equal-weight

#42 meta_equal_weight — DeMiguel-Garlappi-Uppal 2009 1/N over alpha sleeves.

#43meta regime router

Daniel-Moskowitz (2016, JFE) "Momentum Crashes" + Asness-Frazzini- Israel-Moskowitz "Factor Timing" research: different alpha sleeves work in different

Daniel-Moskowitz (2016, JFE) "Momentum Crashes" + Asness-Frazzini-

#13short squeeze meta

Meta-signal building on three components we already ingest: • `stock_borrow_rates` (iborrowdesk / IBKR) → borrow-rate percentile. • `sec_reg_sho_threshold` → Reg-SHO threshold-list membership. • `wikipedia_pageviews` → retail attention z-score (alpha #10 data

Asquith-Pathak-Ritter 2005): when all three fire together, the stock is

short squeeze readiness meta

Combines existing borrow_ftd_squeeze, borrow_rate_spike, ftd_threshold_list into a short-squeeze readiness index that also requires positive attention shock (Wikipedia or social).

Diether-Malloy-Scherbina + Asquith-Pathak-Ritter 2005 (short-sale constraint literature)

Edge: Component-stack edge; cheap win after Wikipedia ingest ships

Want to see them score a real ticker?

Every family above runs nightly against the full universe. Open a stock page and the ones currently firing will sit at the top of the conviction stack.

Open universe See the 9-layer methodology

For informational and educational purposes only. Not financial advice. Learn more