The Risk Museum

Home of Risk Masterpieces

Published trading strategies, tested the way a quant desk would test them — 20 years of data, realistic costs, Newey–West significance, 10,000-draw bootstrap, out-of-sample degradation limits, regime audits, and six-factor decomposition — by an autonomous AI research pipeline that runs every night. The permanent collection: the ones that didn’t survive.

91works tested

49now on view

2held in the vault (survivors)

$920,460 over 22 years$100k in an S&P index fund, same window — the bar every strategy must beat

Curator’s notes: “I named a model after myself. A 1993 model beat it.” · “I brought a neural network. The 1993 model won again.” · “We tightened our own gates. They took the founder’s model first.” · Membership opens with the first exhibits. · The Reading Room — the curator’s shelf. · The Masterpieces Gallery — the anatomy of asymmetry.

The Tulip Wing · mania weather as of 2026-07-07

Haarlem, 1637 — measured, nightly

Names with a two-year run-up ≥ 100% — the screen of Greenwood, Shleifer & You (JFE 2019) run on our public universe. Historically, flagged names went on to crash (≥40% drawdown within two years) 23% of the time vs a 15% matched base rate (1189 episodes since 2005); with acceleration and turnover, 26%.

WDC — two-year run-up +821%

CIEN — two-year run-up +773%

STX — two-year run-up +736%

MU — two-year run-up +609% ✿ hot: acceleration + turnover

FIX — two-year run-up +473%

ECHO — two-year run-up +451%

GLW — two-year run-up +395%

FLEX — two-year run-up +342%

COHR — two-year run-up +330%

WBD — two-year run-up +269%

TPR — two-year run-up +259%

INTC — two-year run-up +257% ✿ hot: acceleration + turnover

Most flagged names do not crash. This is a base-rate elevation on public data, not a prediction and not advice — the museum label says what the exhibit is, nothing more.

Tested 2026-07-09

BRAIN candidate: group_rank(pv13_com_rk_au / ts_mean(pv13_com_rk_au, 120), industry)

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

Tested 2026-07-09

BRAIN candidate: -rank(ts_zscore(diluted_shares_outstanding_adjustment_avg, 250))

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

Tested 2026-07-09

BRAIN candidate: rank(anl4_afv4_div_number / ts_mean(anl4_afv4_div_number, 120))

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

Tested 2026-07-09

BRAIN candidate: rank(ts_delta(historical_volatility_20, 60))

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

Tested 2026-07-09

BRAIN candidate: rank(ts_zscore(implied_volatility_mean_skew_720, 250))

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

Tested 2026-07-09

BRAIN candidate: -rank(ts_zscore(common_stock_buyback_payments, 250))

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

Tested 2026-07-09

BRAIN candidate: group_rank(ts_zscore(anl4_afv4_dts_spe, 250), subindustry)

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

Tested 2026-07-09

BRAIN candidate: group_rank(ts_zscore(anl4_bvps_mean, 250), subindustry)

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

Tested 2026-07-09

BRAIN candidate: group_rank(ts_delta(anl4_capex_mean, 10), subindustry)

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

Tested 2026-07-09

BRAIN candidate: -rank(ts_delta(equity_awards_granted_non_option_period, 5))

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

Tested 2026-07-09

BRAIN candidate: -group_rank(ts_zscore(common_stock_issuance_proceeds, 20), subindustry)

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

Tested 2026-07-09

Heston neural forward-approximator: params->IV surface to <0.5 vol pt with >=100x speedup (Horvath-Muguruza-Tomas program)

Retired at approximation-quality: MAE 66.86 vp / speedup 51684x missed gates

Tested 2026-07-09

Dimopoulos Chain v1.5: neural emissions (MLP on HAR+VIX+GJR+state components) beat the full linear baseline OOS

Retired at vs-linear-frontier-oos: neural emissions did not beat linear ladder (t=-4.37)

Tested 2026-07-09

VRP timing E_no_steamroller: signal-gated SVXY beats always-long SVXY

Retired at excess-vs-always: no significant excess over always-long (t=-0.36, p=0.660)

Tested 2026-07-09

VRP timing B_contango: signal-gated SVXY beats always-long SVXY

Retired at excess-vs-always: no significant excess over always-long (t=0.57, p=0.313)

Tested 2026-07-09

VRP timing C_contango_pinned: signal-gated SVXY beats always-long SVXY

Retired at excess-vs-always: no significant excess over always-long (t=-1.72, p=0.962)

Tested 2026-07-09

VRP timing D_pinned: signal-gated SVXY beats always-long SVXY

Retired at excess-vs-always: no significant excess over always-long (t=-1.68, p=0.959)

Tested 2026-07-09

GEX-scaled position sizing (0.5x exposed -> 1.5x pinned) beats plain vol-targeted SPY after costs

Retired at overlay-excess: overlay excess vs plain vol targeting: t=-3.80, p=1.000

Tested 2026-07-08

Dimopoulos Chain v1.1: GEX x RV states add next-day range info beyond the combined HAR+VIX+GARCH baseline (OOS, monthly-refit)

Retired at beyond-full-baseline-oos: Passed vs HAR+VIX+plain-GARCH (t_full=2.83) but failed the stricter GJR-augmented baseline (t_full=1.02) in the same-day rerun; duplicate-claim guard prevented overwrite. Demoted 2026-07-10 — the strictest run is the ver

Tested 2026-07-08

Walk-forward-tuned TSMOM family (216 variants, 5y/1y, DD<-35% filter, max-Sharpe selection) beats B&H out-of-sample

Retired at oos-excess: stitched OOS excess vs B&H: t=-0.62, p=0.738

Tested 2026-07-08

TSMOM vol-targeted SPY (video 'winner' TREND_MOMENTUM_PRO; Moskowitz 2012 lineage) beats buy-and-hold risk-adjusted across regimes

Retired at excess-significance: excess over B&H is a coin flip (t=0.14, p=0.45); DD reduction real but modest; 9000-attempt source deflator applies

Tested 2026-07-08

Circular/servitization quality: firms disclosing longer equipment useful lives and larger warranty/service obligations are durable-asset custodians with underpriced earnings quality (Stahel stocks-vs-flows via Rishel/Circudyne)

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT

Tested 2026-07-08

Local-news coverage spikes in small caps precede abnormal returns (GDELT volume, v1 all-outlets)

Retired at event-study: 10d CAR t=-1.3 below 2.5 (upper bound)

Tested 2026-07-08

BRAIN candidate: group_rank(fnd6_cptnewqv1300_saleq / ts_mean(fnd6_cptnewqv1300_saleq, 120), industry)

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

Tested 2026-07-08

BRAIN candidate: rank(fnd6_dilavx / ts_mean(fnd6_dilavx, 120))

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

Tested 2026-07-08

BRAIN candidate: group_rank(anl4_capex_mean / ts_mean(anl4_capex_mean, 120), industry)

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

Tested 2026-07-08

BRAIN candidate: -group_rank(ts_zscore(actual_sales_value_annual, 250), sector)

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

Tested 2026-07-08

BRAIN candidate: rank(anl4_afv4_div_low / ts_mean(anl4_afv4_div_low, 120))

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

Tested 2026-07-08

BRAIN candidate: rank(ts_zscore(fnd6_fatp, 250))

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

Tested 2026-07-08

BRAIN candidate: rank(ts_zscore(fnd6_dd2, 250))

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

Tested 2026-07-08

BRAIN candidate: rank(ts_delta(implied_volatility_mean_skew_30, 10))

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

Tested 2026-07-08

BRAIN candidate: rank(ts_delta(anl4_af_cfps_value, 10))

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

Tested 2026-07-08

BRAIN candidate: group_rank(ts_delta(pv13_ustomergraphrank_hub_rank, 10), subindustry)

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

Tested 2026-07-08

BRAIN candidate: rank(snt_buzz / ts_mean(snt_buzz, 60))

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

Tested 2026-07-08

BRAIN candidate: group_rank validation test

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE

Tested 2026-07-08

BRAIN candidate: -rank(ts_delta(implied_volatility_mean_skew_120, 20))

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

Tested 2026-07-08

BRAIN candidate: -rank(ts_zscore(anl4_afv4_div_low, 120))

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

Tested 2026-07-08

BRAIN candidate: -rank(ts_delta(fnd6_txtubend, 20))

Retired at brain-checks: LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

Tested 2026-07-07

Illiquid stocks earn a premium (Amihud 2002)

Retired at validation: IS/OOS degradation 53% > 30% (IS 1.47 -> OOS 0.69) — overfitting signature

Sharpe 1.188max DD -9%$100k → $311,748 over 21y

Tested 2026-07-07

1-month short-term reversal, retested on the ~500-name S&P universe (Jegadeesh 1990)

Retired at validation: Newey-West t=0.95 < 2.5; bootstrap P(Sharpe<=0)=0.168 > 0.05

Sharpe 0.192max DD -17%$100k → $128,773 over 21y

Tested 2026-07-07

12-1 momentum, retested on the ~500-name S&P universe (Jegadeesh & Titman 1993)

Retired at validation: Newey-West t=-0.98 < 2.5; bootstrap P(Sharpe<=0)=0.798 > 0.05; in-sample Sharpe -0.29 <= 0

Sharpe -0.21max DD -51%$100k → $61,519 over 20y

Tested 2026-07-07

Unusually high recent volume predicts higher returns (Gervais, Kaniel, Mingelgrin 2001)

Retired at validation: Newey-West t=-2.31 < 2.5; bootstrap P(Sharpe<=0)=0.987 > 0.05; in-sample Sharpe -0.67 <= 0

Sharpe -0.493max DD -43%$100k → $64,849 over 21y

Tested 2026-07-07

Negative return skewness predicts higher returns (skewness preference / lottery overpricing)

Retired at validation: Newey-West t=-0.87 < 2.5; bootstrap P(Sharpe<=0)=0.829 > 0.05; in-sample Sharpe -0.09 <= 0

Sharpe -0.168max DD -19%$100k → $84,789 over 21y

Tested 2026-07-07

Low lottery demand (small max daily gain) predicts higher returns (Bali, Cakici, Whitelaw 2011 MAX effect)

Retired at validation: Newey-West t=-3.58 < 2.5; bootstrap P(Sharpe<=0)=1.000 > 0.05; in-sample Sharpe -0.65 <= 0

Sharpe -0.699max DD -77%$100k → $23,572 over 21y

Tested 2026-07-07

Low idiosyncratic volatility predicts higher returns (Ang, Hodrick, Xing, Zhang 2006)

Retired at validation: Newey-West t=-3.89 < 2.5; bootstrap P(Sharpe<=0)=1.000 > 0.05; in-sample Sharpe -0.74 <= 0

Sharpe -0.795max DD -77%$100k → $22,966 over 21y

Tested 2026-07-07

Stocks with high trend consistency (share of up days) exhibit continuation — continuous information momentum (Da, Gurun, Warachka 2014 'frog in the pan')

Retired at validation: Newey-West t=-1.77 < 2.5; bootstrap P(Sharpe<=0)=0.948 > 0.05; in-sample Sharpe -0.34 <= 0

Sharpe -0.382max DD -53%$100k → $53,648 over 21y

Tested 2026-07-07

Proximity to 52-week high predicts returns (George & Hwang 2004)

Retired at validation: Newey-West t=-2.22 < 2.5; bootstrap P(Sharpe<=0)=0.988 > 0.05; in-sample Sharpe -0.50 <= 0

Sharpe -0.484max DD -76%$100k → $28,633 over 20y

Tested 2026-07-07

1-month short-term reversal predicts returns (Jegadeesh 1990)

Retired at validation: Newey-West t=0.58 < 2.5; bootstrap P(Sharpe<=0)=0.286 > 0.05; IS/OOS degradation 71% > 30% (IS 0.17 -> OOS 0.05) — overfitting signature

Sharpe 0.125max DD -19%$100k → $116,411 over 21y

Tested 2026-07-07

12-1 month cross-sectional momentum predicts returns (Jegadeesh & Titman 1993)

Retired at validation: Newey-West t=0.09 < 2.5; bootstrap P(Sharpe<=0)=0.420 > 0.05; in-sample Sharpe -0.02 <= 0

Sharpe 0.02max DD -42%$100k → $93,380 over 20y

Verdicts publish on a 3-day lag. Strategies that pass all six stages are not published. Rejection here means the claim failed as we specified and tested it, net of assumed costs, on large-cap US equities — not that the original paper is wrong in every setting.

Open collection: every exhibit is downloadable as machine-readable data — collection.json. Study it, remix it, prove us wrong. (After the Rijksmuseum’s Rijksstudio.)