The Academic Audit

We re-ran the famous published quant factors — Betting-Against-Beta, low-volatility, the 52-week-high effect, book-to-market value — on a survivorship-free, cost-aware 2005–2026 panel. Most of them failed.

• 51 failed after costs
• 4 validated factor-legs
• 2 provisional

We re-ran the famous published quant factors. Most of them failed.

We took well-known strategies from the academic and practitioner literature — the kind catalogued on QuantPedia and QuantConnect, each tracing back to a published paper — and tested them the honest way: on a survivorship-free US equity panel (delisted names such as SIVBQ, FRCB, TWTR, ATVI and SGEN retained; ETFs/ETNs and leveraged/inverse products removed), with modelled, liquidity-aware costs, over a common 2005–2026 out-of-sample window.

The result is deflationary. Most of the celebrated cross-sectional factors — Betting-Against-Beta, the low-volatility effect, the 52-week-high effect, book-to-market value — turned net-negative once costs and delisted losers were put back in. Of the 57 strategies we could adjudicate, only 4 survived — and each only as a diversifying, market-neutral factor-leg. Every survivor loses money in its worst year. None is a standalone strategy.

We tested and adjudicated 57 published academic/QuantPedia strategies on a survivorship-free 2005-2026 US/FX/ETF panel with modelled costs (incl. short interest acquired from FINRA's public API). 51 failed; 4 survived as validated market-neutral factor-legs (all diversifiers with a losing worst year - none standalone); 2 remain provisional. A further 9 strategies are withheld because we could neither confirm nor refute them (missing data / no qualifying securities).

How to read this

This audit is deliberately unglamorous. Before reading any result, read these caveats — they change what the numbers mean.

• DOLLAR FIGURES ARE NOT RETURNS: every net_notional_21y is a 21-year gross-notional P&L sum at fixed-$-per-position sizing (the SAME convention that makes trend-following look like $948M). It is not a capital return and not annualized. Judge strategies by RF, placebo percentile and worst-year RF — never by the dollar figure.
• ALL genuine legs have a NEGATIVE worst-year RF (-0.65 to -0.81): each loses money in its worst year and is only useful COMBINED in a diversified book. None is a standalone tradeable strategy.
• Long/short factors are market-neutral; buy&hold is NOT their benchmark and is reported only as context for long-only/beta strategies.
• Costs are MODELLED (liquidity-aware spread by $ADV + commission), not live fills; no short-leg borrow/locate or market impact.
• Window is 2005-2026 only — post-publication, out-of-sample vs the source papers (which start earlier). A factor failing here does not refute the paper; a factor passing is regime-conditional.
• Placebo pass line (~P95) is within basket-sampling noise: marginal genuines sit within a few baskets of rejected items.

What "validated" means here: the strategy clears three independent bars — a survivorship-free/immune panel, a placebo rank-skill test at or above the 95th percentile (its real net beats 95%+ of random same-size baskets), and cost-aware net — *and* survives adversarial refutation. It is still only a diversifying factor-leg with a losing worst year, not a deployable book on its own.

Validated factor-legs (4)

• Consistent Momentum Strategy: momentum restricted to consistent-return stocks clears placebo P100 as real rank-skill, but is a crash-prone, long-leg-driven diversifier (worst-year RF -0.90).
• Earnings Quality Factor: low-accrual long/short survives survivorship-free + placebo P100 on point-in-time fundamentals; genuine one-sided rank-skill, crash-prone (worst-year RF -0.65).
• FX Carry Trade: classic carry clears placebo P97 as a real risk premium, but marginal (RF 0.99) and concentrated in the 2022–2026 rate regime; a textbook crowded crash trade.
• Size Factor - Small Capitalization Stocks Premium: the size premium survives placebo P100, but measured only on liquid large/mid names, not the microcap tail where the textbook premium actually lives.

Provisional (2)

• Momentum Factor Combined with Asset Growth Effect: passes placebo P100 but is graded SLEEVE_ONLY, very crash-prone and year-concentrated (worst-year RF -0.96).
• R&D Expenditures and Stock Returns: passes placebo P100 but is graded SLEEVE_ONLY, the most crash-prone of the set (worst-year RF -0.97) and rebalance-fragile.

Failed (51)

• Leveraged Etfs Systematic Risk Mgmt: a market-timing overlay with no rank-skill test (worst-year RF -0.96); beta, not alpha.
• Momentum and Reversal Combined with Volatility Effect in Stocks: has real rank-skill (placebo P99.5) but RF is only 0.94 and net far below passive; statistically significant yet economically uninvestable.
• Momentum Effect in Stocks in Small Portfolios: a real cross-sectional signal that is structurally tail-fragile: fails the Rule-26 jackpot test and carries the worst worst-year (-0.87) from an extreme 10-name long/short concentration.
• Paired Switching: thin placebo margin (P96), only 84 trades and a single 2016 jackpot year; too fragile to call genuine.
• Qp Market Seasonality Sell In May: a real calendar seasonal, but simply half-year reduced-beta market exposure (worst-year RF -1.0), not market-neutral alpha.
• Pre-Holiday Effect: the drift is real and survives cost stress but nets only ~$760/yr; economically negligible, not tradeable standalone.
• Asset Class Trend-Following: SPY-only trend rule; passive buy&hold beats it 4x (worst-year RF -1.0); reduced beta/timing, not alpha.
• Qpqc Hp Lowfreq Momentum: 5 of 6 FX pairs are net-negative; only USDCHF passes, and only on 9 trades; a multiple-testing artifact.
• Qpqc Risk Premia Fx: the 4-pair FX portfolio is net-negative; only USDJPY passes in isolation; a single-pair cherry-pick.
• CAPM Alpha Ranking Strategy on Dow 30 Companies: within-universe rank-skill is real (P97), but the current-Dow-30 universe is survivorship-biased and cannot be confirmed genuine without point-in-time membership.
• Combining Smart Factors Momentum and Market Portfolio: contains an explicit long market-portfolio leg, so it is not market-neutral; its return is dominated by equity beta.
• Industry Momentum - Riding Industry Bubbles: long-only (carries full equity beta) plus a partial-2026 stub contributing ~23% of P&L; a beta-plus-jackpot result, not a clean factor-leg.
• Rotation Strategy for SPY, EEM, EFA, TLT, and GLD: no rank-skill beyond the ETF basket itself (placebo P49); diversified beta, not alpha.
• Trend-following Effect in Stocks: the headline $948M is an artifact of fixed-$-per-position sizing; passive buy&hold beats it 9x and the selection placebo is only P96; beta/timing dominated.
• Asset Class Momentum Rotational System: placebo P58 (no rank-skill) and an equal-weight hold of all seven ETFs beats it; the momentum ranking adds nothing over beta.
• Pairs Trading With Country Etfs: RF 0.04 (flat) with max drawdown far above lifetime net; cost-fatal, the four-fill round-trip eats the spread.
• Short Interest Effect Long Short Version: the long/short is net-negative on FINRA data because the short-crowded leg was run over in the 2019–2026 meme-squeeze era.
• 12 Month Cycle in Cross-Section of Stocks Returns: annual return-seasonality ranking is net-negative (placebo P10); no tradeable edge after costs.
• 52-Weeks High Effect in Stocks: ranks at placebo P0 and turns net-negative after costs and delistings; a famous factor that is not free alpha in naive decile form.
• Accrual Anomaly: Sloan accruals long/short is net-negative on the survivorship-free panel; post-publication decay, no edge after costs.
• Asset Growth Effect: net-negative despite marginal rank-skill (placebo P92); the effect does not survive costs.
• Betting Against Beta Factor In Country Equity Indexes: net-negative at placebo P1; the same beta-bleed problem as the single-stock version.
• Betting Against Beta Factor in Stocks: placebo P0 and structurally short market beta, so it bleeds the equity premium every bull year; the famous factor is not investable in naive decile form.
• Combining Fundamental FSCORE and Equity Short-Term Reversals: a textbook survivorship artifact: positive on the biased panel, net-negative once delisted losers are included.
• Currency Momentum Factor: net-negative at placebo P40; no robust cross-sectional currency-momentum edge in our G10 spot data.
• Earnings Announcement Premium: real rank-skill (placebo P100) but the long/short net is negative after costs; cost-fatal on the required turnover.
• Fomc Meeting Effect Primary: the pre-FOMC drift on SPY fails the screen gate; the documented drift does not clear costs as ported.
• Fomc Meeting Effect Variant: the alternative pre-FOMC window also fails the gate; no robust tradeable FOMC-window edge net of costs.
• G-Score Investing: net-negative at placebo P1; no growth-quality edge after costs and delistings.
• Low Volatility Factor Effect in Stocks: placebo P0; the naive decile form is dominated by the short-high-vol leg and fails badly after costs.
• Momentum Factor And Style Rotation Effect: net-negative at placebo P23; no rank-skill beyond noise.
• Momentum Factor Effect In Stocks: plain 12-1 momentum is roughly flat after costs and delistings; the canonical factor barely clears zero in naive decile form.
• Net Payout Yield Effect: net-negative at placebo P15; no payout edge after costs.
• Option-Expiration Week Effect: the calendar effect on SPY fails the screen gate; no tradeable net after costs.
• Pairs Trading with Stocks: net-negative at placebo P52; the classic edge is arbitraged away and cost-fatal on modern spreads.
• Post-Earnings Announcement Effect: real rank-skill (placebo P94) but small-cap-concentrated and net-negative after costs; not tradeable at liquid-ADV scale.
• Qc Dynamic Breakout Ii: an adaptive channel-breakout on ES fails the screen gate; a technical rule with no persistent edge after costs.
• Qc Ichimoku Energy: Ichimoku signals on XLE fail the screen gate; no edge over holding the sector.
• Intraday ETF Momentum: RF -1.0 with 0/14 positive years; the last-half-hour effect does not survive costs on SPY.
• Qp Intraday Seasonality Bitcoin: time-of-day seasonality fails the screen gate after taker fees; no robust hour-of-day edge net of crypto costs.
• Qpqc Dual Thrust: an intraday range-breakout on ES fails the screen gate; a well-known technical rule with no persistent post-cost edge.
• Residual Momentum Factor: net-negative at placebo P0; residualization removes exactly the part that carried plain momentum.
• Reversal During Earnings-Announcements: net-negative at placebo P27; dominated by cost and adverse selection.
• Reversal in Post-Earnings Announcement Drift: real gross rank-skill (placebo P96) but net-negative on the two-day round-trip; cost-fatal.
• ROA Effect within Stocks: net-negative at placebo P54; gross alpha near zero in our point-in-time data, no edge after costs.
• Sector Momentum Rotational System: no rank-skill beyond the sector basket (placebo P46); diversified beta, not alpha.
• Short Term Reversal In Stocks: a real gross effect destroyed by the monthly turnover it requires; net-negative after costs.
• Value (Book-to-Market) Factor: book-to-market (HML) is net-negative; the value premium is absent or negative over our 2005–2026 sample in naive decile form.
• Gross Profitability (GP/assets): real market-neutral rank-skill (beats 99% of random baskets, placebo P99) but risk-adjusted RF is only 0.56; statistically real yet too weak to trade after costs.
• Net Share Issuance: long/short net-negative (RF -0.38); the short-the-issuers leg was run over in the 2005–2026 bull market, the same failure mode as short interest.
• Piotroski F-Score: long/short net-negative (RF -0.43), and the canonical long-only version is pure equity beta at placebo P0 (random selection does better); no selection skill on the broad panel.

A further 9 strategies are withheld because we could neither confirm nor refute them — 8 lack a required data feed (never tested) and 1 (Graham net-nets) has no qualifying securities in the modern liquid universe. We do not publish verdicts we cannot stand behind.