Vasilios Dimopoulos | The Risk Museum, Risk & Quantitative Analysis

I rebuild the data layer when the vendor’s is wrong

An EDGAR filing panel looked complete and was truncated per company at a date that depended on filing rate, the very variable under test. Rebuilding it from the archived shards took the panel from 50,909 to 129,911 rows and episode measurability from 40% to 95%. The bias was invisible in every summary statistic.

I pin every input, then try to break the pin

A study of mine drifted t = −1.80 to −0.89 with no code change, because one input still read a nightly-refreshed cache. Now every input carries a declared as-of, and the fix is verified by running the study with the live readers disabled and diffing the output byte for byte.

My tests refuse to publish

Each study asserts its own leg sign, entry lag and cost model before it is allowed to adjudicate, and stops rather than print a number. Those assertions are mutation-tested, deliberately broken to confirm they fail, because a check that cannot fail is decoration.

Selected work

Four pieces that show judgement rather than throughput. The full catalogue of 281 retired strategies is below.

The published factor zoo does not survive an honest bar

Tested end to end here, 158 of Microsoft Qlib’s technical alphas, 32 within-sector long/short books and a family of calendar effects each returned the same verdict: zero survivors once the significance hurdle is deflated for multiple testing and returns are re-run under a margin of safety.

What it shows · A view held against the consensus, and the willingness to publish a null instead of mining until something passes. See the collection →

I test my own hypotheses to destruction, then audit the test

A bubble-attention hypothesis was tested six ways (regulatory filing intensity, turnover, and Wikipedia pageviews as a direct attention proxy), four of them at balanced power. All rejected, the null holding in the mean, in crash frequency and in the drawdown tail. The reproducibility bug that turned up mid-study was my own, and is written up as a standing lesson rather than quietly patched.

What it shows · Adversarial review of my own result, and multiplicity charged honestly rather than after the fact. See tonight’s bubble watch →

Ten desks behave like three

What a book of many strategies actually buys you, once correlation between them is measured rather than assumed: the difference between nominal and effective diversification, priced with a real risk budget.

What it shows · Portfolio construction and risk budgeting, the part of the job that survives a regime change. See the Trading Floor →

Forecasts priced in advance, scored in public

14 macro forecasts with the probability, the resolution criteria and the source all fixed before the outcome is known, then scored as they resolve, including the ones that go against me.

What it shows · Calibration under scrutiny, and a decision record that cannot be edited after the fact. See the Forecast Wing →

New here? Start with the paper that explains this whole record, browse the research notes, or go straight to recent updates · tonight’s bubble watch, the stocks with elevated crash odds · the collection of tested & retired strategies.

503ideas filed

281retired, on view

2in the vault (survivors)

$943,744 over 22 years$100k in an S&P index fund, same window. The bar every strategy must beat

Of 503 ideas filed, 285 carry a final verdict. 281 are retired and shown below, 2 held in the vault, and 1 still inside the 3-day publication lag. Of the rest, 9 are queued for testing and 209 could not run at all. These are written by machine candidates that call for paid data this free stack does not have. Counting those as “in testing” would flatter the number, so they are shown separately.

What this record is meant to show

A view, held against the consensus. The published factor zoo does not survive an honest bar. Tested end to end here, 158 of Microsoft Qlib’s technical alphas, 32 within-sector long/short books, and a family of calendar effects each returned the same verdict: zero survivors once the significance hurdle is deflated for multiple testing and the returns are re-run under a margin of safety. The street keeps mining price. The base layer this machine has built is the opposite bet. Durable edge lives in positioning and flow, not price patterns, and that most of the work is the discipline to tell the two apart.

A process, with the ego removed. Every rejected idea stays on the wall with the reason it fell short. The gates never loosen, because a loosened gate manufactures a false survivor. And the first model retired under those gates carried my own name: the Dimopoulos Chain, killed across four versions and posted in full. Knowing what you do not yet know, and attacking it in the open, is the entire point of the collection.

Selected working papers

The Factor Zoo Meets an Honest Bar. 158 published alphas, 29 look significant, zero survive a deflated, cost-stressed test.
Measuring the Short Side. a fresh FINRA short-volume panel through the gates: crowding as risk, not signal, and survivorship-flagged.
The Uncorrelated Book. what a 2.35× diversification ratio actually buys a multi-manager platform.
The AI-CDO Is Already Being Built. securitization risk in AI-infrastructure credit, read before the fact.
The Dimopoulos Chain. a four-version post-mortem of a signal that carried my own name, retired by my own gates.
All research notes →

Museum ledger, the last 10 updates (click to expand)

2026-07-26 Working Paper No. 16, and the first study run in a market that is not equities: "Priced Wrong on Purpose." I pre-registered three claims about prediction-market pricing BEFORE the 33.5 GiB dataset finished downloading, then tested them on 250,858 resolved Kalshi markets across 59,748 distinct events. Two of the three failed. Prices are not calibrated, the favourite-longshot bias is confirmed and steep (contracts at 0.30-0.40 resolve YES 17% of the time; contracts at 0.60-0.70 resolve 88%), and my crowding hypothesis was falsified IN SIGN, since one-sided books turn out better calibrated than balanced ones, not worse. The bias survives one-market-per-event and every duration bucket. NOT promoted: there is no order book in the data, so a twenty-point gap on traded prices stays a measurement until the spread is shown not to eat it

2026-07-26 A paid data vendor, tested and rejected before paying. The one open question from Working Paper 16 is whether the spread consumes the calibration gap, which needs historical top-of-book. I took the free tier of a commercial prediction-market API and found it returns zero snapshots for every market in the sample and for a currently-live market too, so its history does not reach the period studied. The question stays open, honestly, and it cost nothing to learn

2026-07-26 A bug in my own factor decomposition, found and fixed. decompose() subtracted the risk-free rate from the backtester's DOLLAR-NEUTRAL long/short return, which already nets its own funding, injecting a spurious -1.76%/yr drift into every alpha estimate. Demonstrated on a pure-noise spread with TRUE alpha of zero: the reported t was -2.24 (spuriously significant) and is -1.16 once fixed. In the sector-exclusion study the same fix removed nine false positives out of eleven. Blast radius checked before claiming impact: ZERO hypotheses had been rejected at the factor stage, so no verdict ever changed, and the bias made the gate stricter rather than looser

2026-07-26 Sector-exclusion tilts, tested: excluding a sector is a factor bet, not a free lunch. Across eleven exclusions the active return moves under 1%/yr in every case, factor R-squared reaches 0.61 for staples and 0.59 for financials, and after FF5+Mom only two carry residual alpha, both NEGATIVE (ex-technology -0.59%/yr t=-2.50, ex-consumer-discretionary -0.36%/yr t=-2.66). Leaving those sectors out cost real money beyond the factor exposure. Tested, shelved

2026-07-26 New engine primitive: Fama-MacBeth cross-sectional panel regression. The pipeline could validate one signal at a time but could not ask which characteristics are priced holding the others constant. First run (5,170 dates, 348 names/day, gross of costs): momentum is NOT priced against controls (t=1.22), while short-term reversal is the strongest characteristic (t=3.29) with a size effect (t=-2.86) and a volatility premium (t=2.78). That independently reproduces the 158-alpha verdict by a different method, since reversal is exactly what doubled costs destroy

2026-07-26 Does talking about AI predict returns? No. I scored all 14,155 harvested EDGAR filings for AI vocabulary: the language exploded 76-fold (0.01 to 0.76 terms per 10,000 words, 2015-2026) and carries no information (Fama-MacBeth t=1.80 against controls), while a long-high/short-low book LOSES money and fails every gate. Declared limit: term counting cannot separate deploying AI from disclosing AI as a threat in risk factors

2026-07-26 Research note #9 gains a primary-source section. Three Stanford MS&E 435 lectures (Spring 2026) put the people doing the spending on the record, and I pulled the transcripts directly: OpenAI's compute lead prices a gigawatt at $70 billion and half a million GPUs, against an aspirational 30 GW target and ~100 GW of US hyperscaler plans. He names ASML as the single chokepoint of the whole supply chain, which is the hidden-correlation problem stated by the buyer, and describes a three-year chip design cycle against a four-to-six-year depreciation schedule. Across three hours of senior operators, debt is never mentioned once. Attributed, not endorsed

2026-07-26 Working Paper No. 15, the overnight capstone: "Lazy Prices Meets Survivorship." I harvested 14,155 EDGAR 10-Q/10-K filings for 406 names and tested the disclosure-text anomaly (rewrite your filing, underperform; Cohen-Malloy-Nguyen 2020). It does not replicate: the year-over-year text-change measure is specification-fragile and, cleaned, turns significant but OPPOSITE-signed (|t|=2.55), below the bar. The cause is survivorship, a current-constituents panel deletes the informative short leg, the firms that rewrote their disclosure on the way to trouble. A published edge, disqualified by the data I have. The vault holds at two, no manufactured win

2026-07-26 Market concentration, tested as a regime, the narrow-market question with numbers. The top-10 share of universe dollar volume runs 18-50% (43% now), and equal-weight lags cap-weight by -3.3%/yr when the market is top-heavy versus -0.75%/yr when it is broad, the Mag-7 dynamic made explicit. But a regime-timed equal-weight/cap-weight tilt fails the gates (t=-0.68). Like gold: a real regime, not a tradable signal. Tested, shelved

2026-07-26 Gold, tested as a conditional diversifier, the hedge-fund macro question, answered with numbers. Over 21 years gold returned 11.7%/yr (Sharpe 0.64), and MORE in the positive stock-bond-correlation regime (19.3% vs 9.8%), the "bonds stopped hedging" world of 2023 on. But its actual vol-reduction for a 60/40 book is roughly regime-invariant (~0.4-0.6%, slightly better when bonds DO hedge), and a regime-timed gold overlay fails the gates (t=-1.04). Gold is a real diversifier, not a tradable regime signal. Tested, shelved

Stocks that double in two years go on to crash more often. These just did.

The Tulip Wing · a nightly bubble watch · watch list updated August 21, 2026 · statistics frozen at 2026-07-29

Why this list matters: since 2005, a stock that doubled inside two years went on to crash (a 40%+ fall within two years) 23% of the time, versus 15% for a control group of names drawn at random from the same months. When the rise is accelerating on heavy trading, 26%. A doubling is not a death sentence. It is a measured tilt in the odds, and these are the names carrying it tonight:

The honest sample size, and what it does to this claim. The 23% above counts every month a name sits in tulip territory, so one long mania is counted many times. Counted once per episode instead, which is the honest unit and the one my nightly report has always flagged, the crash rate is 20% against the same 15% control. That is 1.28x the control rate against 1.51x on the inflated count, and my own significance bar for this screen is 1.25. So on the honest count it clears that bar, but only just, and it sits close enough to that bar that an earlier version of this screen straddled it, measuring 1.28 on one nightly pull and 1.18 on another as the universe moved. A result whose pass or fail depends on the data fetch is not a result, so every statistic in this paragraph now comes from a frozen panel of 437 names through 2026-07-29, prices and trading volume both. The verdict is pinned and reproducible; only the watch list below is live. I still would not defend a margin this thin hard. The control is also matched on the calendar only, not on volatility or sector, which the source paper matched on.

WDC, two-year run-up +853%

MU, two-year run-up +801%

STX, two-year run-up +766%

CIEN, two-year run-up +636%

FIX, two-year run-up +404%

ECHO, two-year run-up +384%

INTC, two-year run-up +332%

GLW, two-year run-up +286%

WBD, two-year run-up +285%

COHR, two-year run-up +278%

LRCX, two-year run-up +267%

FLEX, two-year run-up +249%

The method is the Harvard bubble screen (Greenwood, Shleifer & You, 2019): a two-year run-up of 100% or more, checked nightly against my universe. The wing is named for tulip mania: in 1637, at a bulb auction in Haarlem, buyers simply stopped bidding, and history’s most famous bubble collapsed.

Read the numbers again: most flagged names do not crash. This is a measured tilt in the odds, not a prediction and not advice. The museum label says what the exhibit is, nothing more.

Tested strategies: the permanent collection

281 strategies tested and retired · newest first

Each plaque records the claim and the gate that retired it. The strategies that reached a full backtest are totalled first, the way a desk reads its book. What $100k put into each would have become.

The backtested wall · $100k in, sorted best to worst

Strategy	Style	Sharpe	Max DD	Yrs	$100k →	Net P&L	Return
S&P 500 index fund Buy and hold, the bar every strategy must clear same 22-year window	index	n/a	n/a	22	$943,744	+$843,744	+844%
Illiquid stocks earn a premium IS/OOS degradation 53% > 30% (IS 1.47 -> OOS 0.69), overfitting signature tested 2026-07-07 · retired at validation	Liquidity	1.19	−9%	21	$311,748	+$211,748	+212%
1-month short-term reversal, retested on the ~500-n… Newey-West t=0.95 < 2.5; bootstrap P(Sharpe<=0)=0.168 > 0.05 tested 2026-07-07 · retired at validation	Reversal	0.19	−17%	21	$128,773	+$28,773	+29%
1-month short-term reversal predicts returns 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)… tested 2026-07-07 · retired at validation	Reversal	0.12	−19%	21	$116,411	+$16,411	+16%
Slow-spread pairs Newey-West t=0.47 < 2.74 (base 2.5, family n=2); bootstrap P(Sharpe<=0)=0.3692 > 0.025 (base 0.05, family n=2);… tested 2026-08-11 · retired at validation	Cross-sectional	0.10	−25%	20	$109,181	+$9,181	+9%
Market-residual pairs Newey-West t=0.42 < 2.87 (base 2.5, family n=3); bootstrap P(Sharpe<=0)=0.3825 > 0.01667 (base 0.05, family n=3);… tested 2026-08-11 · retired at validation	Cross-sectional	0.09	−15%	20	$105,832	+$5,832	+6%
Distance-pairs divergence as a cross-sectional sign… Newey-West t=0.22 < 2.50 (base 2.5, family n=1); bootstrap P(Sharpe<=0)=0.4616 > 0.05 (base 0.05, family n=1);… tested 2026-08-11 · retired at validation	Cross-sectional	0.05	−19%	20	$102,208	+$2,208	+2%
12-1 month cross-sectional momentum predicts returns Newey-West t=0.09 < 2.5; bootstrap P(Sharpe<=0)=0.420 > 0.05; in-sample Sharpe -0.02 <= 0 tested 2026-07-07 · retired at validation	Momentum	0.02	−42%	20	$93,380	−$6,620	−7%
Negative return skewness predicts higher returns Newey-West t=-0.87 < 2.5; bootstrap P(Sharpe<=0)=0.829 > 0.05; in-sample Sharpe -0.09 <= 0 tested 2026-07-07 · retired at validation	Skewness	−0.17	−19%	21	$84,789	−$15,211	−15%
Unusually high recent volume predicts higher returns Newey-West t=-2.31 < 2.5; bootstrap P(Sharpe<=0)=0.987 > 0.05; in-sample Sharpe -0.67 <= 0 tested 2026-07-07 · retired at validation	Volume	−0.49	−43%	21	$64,849	−$35,151	−35%
12-1 momentum, retested on the ~500-name S&P univer… Newey-West t=-0.98 < 2.5; bootstrap P(Sharpe<=0)=0.798 > 0.05; in-sample Sharpe -0.29 <= 0 tested 2026-07-07 · retired at validation	Momentum	−0.21	−51%	20	$61,519	−$38,481	−38%
Stocks with high trend consistency Newey-West t=-1.77 < 2.5; bootstrap P(Sharpe<=0)=0.948 > 0.05; in-sample Sharpe -0.34 <= 0 tested 2026-07-07 · retired at validation	Momentum	−0.38	−53%	21	$53,648	−$46,352	−46%
Proximity to 52-week high predicts returns Newey-West t=-2.22 < 2.5; bootstrap P(Sharpe<=0)=0.988 > 0.05; in-sample Sharpe -0.50 <= 0 tested 2026-07-07 · retired at validation	Momentum	−0.48	−76%	20	$28,633	−$71,367	−71%
Low lottery demand Newey-West t=-3.58 < 2.5; bootstrap P(Sharpe<=0)=1.000 > 0.05; in-sample Sharpe -0.65 <= 0 tested 2026-07-07 · retired at validation	Lottery	−0.70	−77%	21	$23,572	−$76,428	−76%
Low idiosyncratic volatility predicts higher returns Newey-West t=-3.89 < 2.5; bootstrap P(Sharpe<=0)=1.000 > 0.05; in-sample Sharpe -0.74 <= 0 tested 2026-07-07 · retired at validation	Low-risk	−0.80	−77%	21	$22,966	−$77,034	−77%

Not one retired strategy cleared the index, which is why they hang here and not in the vault. Sharpe is an upper bound on a survivorship-biased universe; every figure is a hypothetical backtest net of assumed costs, not a live result. Every strategy here is cross-sectional across US large-caps, long the factor, short its opposite, so it touches every sector by construction; the Style column is the factor family, not a GICS sector.

How to read the collection, including what “BRAIN candidate” and the retirement codes mean

BRAIN candidates are alpha expressions written by machine formulas (in WorldQuant BRAIN syntax) that the swarm screens on four fast platform checks before any full backtest; the plain-English line under each says what the formula actually buys and sells.

Retirement codes. LOW_SHARPE: weak risk-adjusted return · LOW_FITNESS: weak on BRAIN’s blended quality score · LOW_SUB_UNIVERSE_SHARPE: fails on the smaller-stock subset · CONCENTRATED_WEIGHT: the bet piles into too few names · SELF_CORRELATION: it duplicates a signal I already have.

Tested 2026-07-08 to 2026-08-21

204 machine-written alpha expressions, one collective tombstone

Retired at the BRAIN screen: every one of these came off the same generator and died on the platform's own checks, low Sharpe, low fitness and self-correlation above all. They hang together because 204 near-identical stones would bury the rest of the collection.

All 204 expressions with their failure codes (click to expand)

2026-08-21 group_rank(ts_zscore(current_foreign_tax_expense_value, 250), subindustry) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-21 group_rank(ts_zscore(scl12_buzz, 250), subindustry) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-21 rank(ts_delta(snt_buzz_ret, 10)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-20 rank(zscore(anl4_af_div_value) - zscore(anl4_afv4_div_mean)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-20 rank(ts_delta(common_stock_repurchase_payment, 60)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-20 -rank(implied_volatility_mean_1080 / ts_mean(implied_volatility_mean_1080, 120)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-08-19 -rank(ts_zscore(scl12_buzz_fast_d1, 250)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-19 group_zscore(ts_delta(diluted_shares_outstanding_adjustment, 120), industry) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-19 group_zscore(ts_delta(fnd6_cstkcv, 120), industry) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-08-19 rank(anl4_cfo_mean / ts_mean(anl4_cfo_mean, 120)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-19 rank(zscore(anl4_bvps_median) - zscore(common_stock_buyback_payments)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-18 rank(ts_delta(anl4_afv4_dts_spe, 60)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-18 rank(ts_delta(current_federal_tax_expense_amount, 60)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-18 rank(ts_delta(pv13_revere_key_sector_total, 10)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-08-18 group_rank(ts_delta(fnd6_dn, 10), subindustry) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-08-17 -rank(debt_maturities_repayments_next12m / ts_mean(debt_maturities_repayments_next12m, 120)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-17 rank(ts_delta(fnd6_cptnewqv1300_epsx12, 10)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-17 rank(fnd6_cptrank_gvkeymap / ts_mean(fnd6_cptrank_gvkeymap, 120)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-16 -group_rank(ts_zscore(anl4_bvps_median, 250), sector) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-16 rank(ts_corr(common_stock_buyback_payments, fair_value_derivative_liabilities, 120)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-16 rank(zscore(equipment_maximum_useful_life) - zscore(fnd6_fatc)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-16 rank(ts_delta(anl4_afv4_dts_spe, 10)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-16 rank(zscore(implied_volatility_mean_skew_20) - zscore(fnd6_dd2)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-08-15 group_rank(ts_delta(pv13_revere_country, 10), subindustry) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-08-14 -group_rank(ts_zscore(snt_buzz_bfl, 20), subindustry) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-14 -rank(implied_volatility_mean_1080 / ts_mean(implied_volatility_mean_1080, 120)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-08-14 group_zscore(ts_delta(actual_sales_value_quarterly, 120), industry) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-13 rank(ts_corr(common_stock_issuance_proceeds_2, returns, 20)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-08-13 rank(zscore(fnd6_aqs) - zscore(rel_num_cust)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-13 -rank(business_acquisition_payments_net / ts_mean(business_acquisition_payments_net, 120)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-12 -rank(ts_zscore(current_state_local_tax_expense_amount, 250)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-12 group_rank(zscore(parkinson_volatility_60) - zscore(pv13_revere_level), subindustry) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-08-12 group_zscore(ts_delta(common_stock_issuance_proceeds_2, 120), industry) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-12 group_rank(ts_zscore(scl12_buzz, 250), subindustry) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-12 rank(ts_zscore(finite_intangibles_gross_value, 250)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-12 group_rank(ts_zscore(parkinson_volatility_10, 250), subindustry) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-08-12 rank(ts_delta(parkinson_volatility_30, 60)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-08-11 group_rank(zscore(snt_value) - zscore(credit_facility_outstanding_amount), subindustry) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-11 rank(ts_corr(domestic_ebit_value, returns, 20)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-11 group_rank(ts_delta(deferred_tax_liability_property_plant_equipment, 10), subindustry) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-08-11 group_rank(ts_delta(implied_volatility_put_10, 10), subindustry) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-08-10 group_rank(implied_volatility_mean_60 / ts_mean(implied_volatility_mean_60, 120), industry) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-08-10 group_rank(ts_zscore(anl4_bvps_value, 250), subindustry) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-10 group_rank(comprehensive_income_net_tax / ts_mean(comprehensive_income_net_tax, 120), industry) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-10 group_rank(pv13_revere_parent / ts_mean(pv13_revere_parent, 120), industry) · LOW_SHARPE, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-08-10 -rank(ts_zscore(equity_awards_granted_non_option_period, 250)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-10 -group_rank(ts_zscore(parkinson_volatility_180, 250), sector) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-08-10 group_rank(deferred_tax_liabilities_total_4 / ts_mean(deferred_tax_liabilities_total_4, 120), industry) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-10 rank(anl4_af_eps_value / ts_mean(anl4_af_eps_value, 120)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-10 -rank(deferred_tax_liability_property_plant_equipment / ts_mean(deferred_tax_liability_property_plant_equipment, 120)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-10 rank(ts_delta(anl4_af_eps_value, 10)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-08-10 rank(zscore(implied_volatility_mean_360) - zscore(snt_value)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-08-10 -rank(ts_delta(anl4_afv4_eps_high, 5)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-08-09 rank(ts_delta(common_stock_buyback_payments, 60)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-09 -rank(rel_num_part / ts_mean(rel_num_part, 120)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-09 rank(ts_rank(implied_volatility_put_20, 250)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-08-09 group_rank(zscore(fnd6_beta) - zscore(implied_volatility_mean_skew_90), subindustry) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-08-09 rank(zscore(anl4_afv4_div_number) - zscore(comprehensive_income_net_tax)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-09 rank(ts_delta(current_federal_tax_expense_amount, 60)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-09 -rank(ts_zscore(multi_factor_acceleration_score_derivative, 250)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-09 rank(zscore(fnd6_beta) - zscore(anl4_afv4_eps_high)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-09 -rank(ts_zscore(anl4_afv4_dts_spe, 250)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-09 rank(ts_corr(antidilutive_securities_excluded_eps, returns, 20)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-08-09 rank(zscore(credit_facility_max_borrowing) - zscore(anl4_afv4_div_high)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-09 -group_rank(ts_zscore(anl4_af_div_value, 20), subindustry) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-08-08 rank(zscore(common_stock_issuance_proceeds) - zscore(parkinson_volatility_90)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-08-08 group_rank(zscore(anl4_afv4_cfps_number) - zscore(current_income_tax_expense_amount), subindustry) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-08 rank(ts_rank(rel_num_cust, 250)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-08 rank(ts_rank(fnd6_dxd4, 250)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-08 rank(zscore(actual_eps_value_quarterly) - zscore(anl4_cff_mean)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-08 -rank(ts_zscore(anl4_afv4_median_eps, 250)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-08 -rank(diluted_shares_outstanding_adjustment_avg / ts_mean(diluted_shares_outstanding_adjustment_avg, 120)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-08 -rank(ts_zscore(anl4_afv4_div_median, 250)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-08 -group_rank(ts_zscore(deferred_tax_liabilities_total_4, 250), sector) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-08 rank(ts_delta(finite_intangibles_gross_value, 10)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-08-08 rank(ts_delta(pv13_revere_key_sector_total, 10)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-08-08 -rank(ts_delta(anl4_afv4_eps_high, 5)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-07 group_zscore(ts_delta(rel_num_cust, 120), industry) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-07 rank(ts_rank(fnd6_dxd2, 250)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-07 rank(ts_delta(debt_repayment_year_three, 60)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-07 group_rank(fnd6_fato / ts_mean(fnd6_fato, 120), industry) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-07 rank(zscore(snt_buzz_ret) - zscore(parkinson_volatility_150)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-08-07 group_rank(zscore(debt_repayment_year_three) - zscore(deferred_tax_liability_property_plant_equipment), subindustry) · LOW_SHARPE, LOW_FITNESS, LOW_TURNOVER, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-07 -group_rank(ts_zscore(snt_buzz_fast_d1, 20), subindustry) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-07 rank(ts_delta(current_state_local_tax_expense_amount, 10)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-08-07 group_rank(zscore(credit_facility_outstanding_amount) - zscore(scl12_sentiment_fast_d1), subindustry) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-07 rank(ts_corr(pv13_ustomergraphrank_hub_rank, earnings_certainty_rank_derivative, 120)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-07 -group_rank(ts_zscore(anl4_afv4_div_mean, 250), sector) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-07 group_rank(ts_delta(implied_volatility_mean_20, 10), subindustry) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-08-05 rank(ts_corr(anl4_afv4_median_eps, returns, 20)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-04 -rank(winsorize(ts_zscore(adj_net_income_avg, 500), std=4)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-03 group_zscore(ts_delta(implied_volatility_call_20, 120), industry) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-08-03 rank(zscore(anl4_cfo_mean) - zscore(anl4_cfi_number)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-03 -rank(ts_zscore(common_stock_buyback_payments, 250)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-08-03 -group_rank(ts_zscore(fnd6_dltp, 250), sector) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-03 group_rank(ts_delta(snt_buzz, 10), subindustry) · LOW_SHARPE, LOW_FITNESS, HIGH_TURNOVER, SELF_CORRELATION

2026-08-03 group_rank(ts_delta(anl4_af_cfps_value, 10), subindustry) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-08-03 -rank(fnd6_acqintan / ts_mean(fnd6_acqintan, 120)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-08-03 -rank(ts_zscore(implied_volatility_mean_720, 250)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-08-03 group_zscore(ts_delta(implied_volatility_mean_skew_30, 120), industry) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-08-03 -rank(equipment_maximum_useful_life / ts_mean(equipment_maximum_useful_life, 120)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-08-03 -rank(ts_delta(rel_ret_all, 5)) · LOW_SHARPE, LOW_FITNESS, HIGH_TURNOVER, SELF_CORRELATION

2026-08-03 rank(zscore(implied_volatility_mean_30) - zscore(business_acquisition_payments_net)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-07-29 -rank(ts_zscore(rel_ret_part, 250)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-29 rank(zscore(implied_volatility_mean_skew_10) - zscore(diluted_shares_outstanding_adjustment_avg)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-07-29 rank(anl4_afv4_div_low / ts_mean(anl4_afv4_div_low, 120)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-29 rank(ts_rank(parkinson_volatility_90, 250)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-07-29 group_rank(scl12_buzz_fast_d1 / ts_mean(scl12_buzz_fast_d1, 120), industry) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-29 rank(zscore(pv13_com_page_rank) - zscore(anl4_afv4_dts_spe)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-29 rank(ts_corr(finite_intangibles_gross_value, returns, 20)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-07-29 group_zscore(ts_delta(effective_tax_rate_continuing_ops_2, 120), industry) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-29 rank(ts_rank(finite_intangibles_gross_value, 250)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-29 -rank(anl4_bvps_median / ts_mean(anl4_bvps_median, 120)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-29 rank(ts_delta(parkinson_volatility_180, 60)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-07-29 rank(ts_corr(finite_intangibles_gross_value, returns, 20)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-07-27 rank(ts_delta(domestic_ebit_value, 10)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-27 rank(zscore(adj_net_income_median) - zscore(anl4_af_eps_value)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-26 rank(anl4_cff_mean / ts_mean(anl4_cff_mean, 120)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-25 -rank(winsorize(ts_zscore(equipment_maximum_useful_life, 500), std=4)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-25 rank(ts_delta(anl4_afv4_dts_spe, 60)) · SELF_CORRELATION

2026-07-25 -rank(winsorize(ts_zscore(adj_net_income_median, 500), std=4)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-25 rank(ts_zscore(implied_volatility_mean_20, 250)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-07-25 rank(ts_corr(anl4_bvps_median, exercisable_options_count, 120)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-07-25 group_rank(ts_zscore(business_acquisition_payments_net, 250), subindustry) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-25 -group_rank(ts_zscore(snt_value_fast_d1, 250), sector) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-25 rank(ts_zscore(actual_sales_value_annual, 250)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-25 -group_rank(ts_zscore(anl4_bvps_value, 20), subindustry) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-07-25 -group_rank(ts_zscore(implied_volatility_mean_150, 20), subindustry) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-07-25 -group_rank(ts_zscore(fnd6_dd2, 20), subindustry) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-07-25 -group_rank(ts_zscore(snt_value_fast_d1, 250), sector) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-24 rank(fnd6_cshr / ts_mean(fnd6_cshr, 120)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-24 rank(anl4_afv4_cfps_low / ts_mean(anl4_afv4_cfps_low, 120)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-24 group_zscore(ts_delta(implied_volatility_put_10, 120), industry) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-07-24 rank(ts_zscore(fnd6_beta, 250)) · LOW_SHARPE, LOW_FITNESS, LOW_TURNOVER, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHA

2026-07-24 rank(zscore(anl4_dts_ptp) - zscore(anl4_cff_number)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-24 rank(ts_delta(comprehensive_income_net_tax_value, 10)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-07-24 rank(ts_delta(debt_issuance_proceeds, 60)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-24 group_rank(zscore(pv13_revere_term_sector_total) - zscore(deferred_tax_assets_compensation_benefits), subindustry) · LOW_SHARPE, LOW_FITNESS, LOW_TURNOVER, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-24 rank(ts_rank(annual_intangible_assets_net_carrying_value, 250)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-24 rank(ts_delta(anl4_bvps_median, 10)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-23 -rank(snt_value / ts_mean(snt_value, 120)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-22 rank(ts_delta(implied_volatility_mean_skew_60, 60)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-07-21 rank(fnd6_acqintan / ts_mean(fnd6_acqintan, 120)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-19 -rank(implied_volatility_mean_270 / ts_mean(implied_volatility_mean_270, 120)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-07-18 rank(ts_delta(actual_sales_value_annual, 10)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-18 rank(zscore(anl4_capex_number) - zscore(actual_dividend_value_quarterly)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-17 -group_rank(ts_zscore(anl4_afv4_eps_high, 20), subindustry) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-16 group_zscore(ts_delta(anl4_bvps_mean, 120), industry) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-16 rank(ts_rank(implied_volatility_mean_skew_1080, 250)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-07-16 -rank(winsorize(ts_zscore(anl4_bvps_value, 500), std=4)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-16 -group_rank(ts_zscore(implied_volatility_call_20, 250), sector) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-07-16 -rank(snt_value_fast_d1 / ts_mean(snt_value_fast_d1, 120)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-16 group_zscore(ts_delta(implied_volatility_put_1080, 120), industry) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-07-16 rank(ts_zscore(debt_repayment_year_three, 250)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-16 -group_rank(ts_zscore(actual_sales_value_annual, 250), sector) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-16 rank(ts_delta(fnd6_cptmfmq_actq, 60)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-16 rank(zscore(accumulated_oci_net_of_tax_value) - zscore(deferred_local_income_tax_expense)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-16 -rank(ts_delta(debt_issuance_costs_expense, 5)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-07-16 -rank(ts_delta(current_federal_tax_expense_amount, 5)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-07-15 rank(ts_rank(pv13_custretsig_retsig, 250)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-15 rank(anl4_cfo_number / ts_mean(anl4_cfo_number, 120)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-15 rank(ts_corr(implied_volatility_mean_90, diluted_shares_outstanding_adjustment_avg, 120)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-07-15 group_rank(anl4_af_div_value / ts_mean(anl4_af_div_value, 120), industry) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-15 group_rank(debt_repayments_total_2 / ts_mean(debt_repayments_total_2, 120), industry) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-15 rank(ts_zscore(implied_volatility_mean_1080, 250)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-07-15 group_rank(ts_delta(actual_sales_value_annual, 10), subindustry) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-15 rank(ts_corr(implied_volatility_mean_30, returns, 20)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-15 group_zscore(ts_delta(parkinson_volatility_150, 120), industry) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-07-15 -group_rank(ts_zscore(historical_volatility_30, 250), sector) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-07-15 rank(ts_corr(anl4_cff_mean, snt_buzz_bfl_fast_d1, 120)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-15 -group_rank(ts_zscore(fnd6_fatp, 20), subindustry) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-07-14 -rank(pv13_revere_parent / ts_mean(pv13_revere_parent, 120)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-07-14 rank(ts_delta(historical_volatility_90, 10)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-07-13 rank(ts_zscore(fair_value_derivative_liabilities, 250)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-13 rank(ts_delta(implied_volatility_mean_180, 60)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-07-13 -rank(ts_zscore(anl4_afv4_div_high, 250)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-12 group_rank(zscore(historical_volatility_20) - zscore(anl4_bvps_high), subindustry) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-07-11 -group_rank(ts_zscore(fnd6_dcvsub, 250), sector) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-07-09 group_rank(pv13_com_rk_au / ts_mean(pv13_com_rk_au, 120), industry) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-09 -rank(ts_zscore(diluted_shares_outstanding_adjustment_avg, 250)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-09 rank(anl4_afv4_div_number / ts_mean(anl4_afv4_div_number, 120)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-09 rank(ts_delta(historical_volatility_20, 60)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-07-09 rank(ts_zscore(implied_volatility_mean_skew_720, 250)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-07-09 -rank(ts_zscore(common_stock_buyback_payments, 250)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-09 group_rank(ts_zscore(anl4_afv4_dts_spe, 250), subindustry) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-09 group_rank(ts_zscore(anl4_bvps_mean, 250), subindustry) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-09 group_rank(ts_delta(anl4_capex_mean, 10), subindustry) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-09 -rank(ts_delta(equity_awards_granted_non_option_period, 5)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-07-09 -group_rank(ts_zscore(common_stock_issuance_proceeds, 20), subindustry) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-07-08 group_rank(fnd6_cptnewqv1300_saleq / ts_mean(fnd6_cptnewqv1300_saleq, 120), industry) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-08 rank(fnd6_dilavx / ts_mean(fnd6_dilavx, 120)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-08 group_rank(anl4_capex_mean / ts_mean(anl4_capex_mean, 120), industry) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-08 -group_rank(ts_zscore(actual_sales_value_annual, 250), sector) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-08 rank(anl4_afv4_div_low / ts_mean(anl4_afv4_div_low, 120)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-08 rank(ts_zscore(fnd6_fatp, 250)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-08 rank(ts_zscore(fnd6_dd2, 250)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-08 rank(ts_delta(implied_volatility_mean_skew_30, 10)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

2026-07-08 rank(ts_delta(anl4_af_cfps_value, 10)) · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE, SELF_CORRELATION

2026-07-08 group_rank(ts_delta(pv13_ustomergraphrank_hub_rank, 10), subindustry) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-07-08 rank(snt_buzz / ts_mean(snt_buzz, 60)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-08 group_rank validation test · LOW_SHARPE, LOW_FITNESS, LOW_SUB_UNIVERSE_SHARPE

2026-07-08 -rank(ts_delta(implied_volatility_mean_skew_120, 20)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, SELF_CORRELATION

2026-07-08 -rank(ts_zscore(anl4_afv4_div_low, 120)) · LOW_SHARPE, LOW_FITNESS, SELF_CORRELATION

2026-07-08 -rank(ts_delta(fnd6_txtubend, 20)) · LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT, LOW_SUB_UNIVERSE_SHARPE, SELF_CORR

Tested 2026-08-23

13F shadow of equity held on swap, via dealer entities with no asset-management arm

Synthetic crowding: equity held on swap is invisible to 13F crowding measures, but the hedging dealer's own 13F is not. Measure the share of each name held by derivatives-only dealer entities and test whether the Archegos book was visible in the quarter before the unwind.

Retired at no promotion claim: Measurement exhibit by design: the episode count cannot clear the deflated, cost-stressed bar. Filed so the quantity and its limits are on the record.

Tested 2026-08-15

Event set: Item 1.01 8-Ks whose body matches agreement-and-plan-of-merger, deterministic keyword filter, no model judgment. Point-in-time entry at the first close after acceptance datetime. One declared window, 20 trading days, long acquirer hedged short SPY, engine costs. Honesty: this is the ACQUIRER side only; the classic deal-spread trade needs target prices and targets delist, so the target-side study is declared blocked on the delisted-price vein, second in the constraint audit elevation order. Survivor panel biases the acquirer sample toward firms that stayed large, stated.

Acquirers drift after signing definitive merger agreements: 8-K filings under Item 1.01 whose body contains a definitive merger agreement, timestamped by SEC acceptance datetime, are followed by measurable abnormal returns in the acquirer over the next 20 trading days versus the market. First look of the deal-events family, the one mechanism lane the store has never tested, using the freshly opened 8-K body vein.

Retired at validation: Newey-West t=-3.07 < 2.50 (base 2.5, family n=1); bootstrap P(Sharpe<=0)=0.9992 > 0.05 (base 0.05, family n=1); margin-of-safety rerun (gross halved, costs doubled) t=-3.16 < 2.5, edge has no buffer; in-sample Sharpe …

Tested 2026-08-14

Data: N-PORT net assets are public quarterly from late 2019, so the sample starts there, which also bypasses the v1 cache defect of zero-volume SOXS and TECS days before 2019. Intra-quarter assets scaled by fund NAV return, ignoring flows within the quarter, a declared approximation. v1 died of cost-death on a real gross effect with the proxy shown to be a volatility filter; this look isolates the assets variable.

Levered ETF rebalancing pressure with REAL fund assets: scaling the forced-flow signal by SEC-reported net assets (N-PORT filings, quarterly anchors interpolated intra-quarter by fund NAV returns) instead of the v1 volume proxy isolates the Morgan Stanley QDS variable the first look never tested. Same trade rule as v1, unchanged: top-quintile pressure days, fade the sign, one day lag, engine costs. Judged at the family n=2 bar.

Retired at validation: Newey-West t=0.89 < 2.74 (base 2.5, family n=2); bootstrap P(Sharpe<=0)=0.1933 > 0.025 (base 0.05, family n=2); margin-of-safety rerun (gross halved, costs doubled) t=-1.28 < 2.5, edge has no buffer; IS/OOS degradation …

Tested 2026-08-14

Needs a levered_etf_lab runner: fetch ~12 major levered/inverse ETF price and volume series via the existing Yahoo path (TQQQ SQQQ SOXL SOXS UPRO SPXU SSO SDS QLD QID TECL TECS), construct signed pressure per underlier as (L squared minus L) times AUM-proxy times daily return, test next-day reversal event-study style with the engine standard stats. v1 AUM proxy is fund dollar volume, declared as a proxy; historical AUM acquisition is the upgrade path. Index-level, few assets, so research-lane not cross-sectional. Untested family, first look.

Levered ETF rebalancing pressure predicts next-day reversal in the underlying indexes. The mechanism is forced flow: funds holding L-times exposure must trade (L squared minus L) times AUM times the day's return at the close, same direction as the move, regardless of price. MS QDS: over $200bn AUM, $100bn sold June 5 to July 29 2026, ~$30bn bought on July 30 alone, 2 to 4% of US equity volume on peak days. Predictable, mechanical, and concentrated at the close: the passing lane's shape exactly.

Retired at validation: Newey-West t=-0.03 < 2.50 (base 2.5, family n=1); bootstrap P(Sharpe<=0)=0.5333 > 0.05 (base 0.05, family n=1); margin-of-safety rerun (gross halved, costs doubled) t=-3.14 < 2.5, edge has no buffer; IS/OOS degradation …

Tested 2026-08-11

The uncontaminated version of the slow-spread look. Judged after the headline variant, n=5 at judge time.

Slow-spread pair-level selection: among the 200 closest formation pairs, trading only those whose spread AR(1) half-life is 10 to 60 days preserves net-of-lag convergence profit. Same book construction as the headline pair-level test, fixed anchor so slow divergences are no longer absorbed by refresh re-basing.

Retired at validation: Newey-West t=-0.86 < 3.03 (base 2.5, family n=5); bootstrap P(Sharpe<=0)=0.8183 > 0.01 (base 0.05, family n=5); margin-of-safety rerun (gross halved, costs doubled) t=-1.62 < 2.5, edge has no buffer; in-sample Sharpe …

Tested 2026-08-11

The critic-prescribed fair test after three cross-sectional conversions died. Survivorship inflates convergence for pairs specifically, so any pass is an upper bound. Judged at the family-deflated bar, n=4 at judge time.

Pair-level GGR pairs trading, run faithfully on large-cap survivors: top 20 mutual min-SSD pairs, 252d formation with fixed anchor, 126d trading, open at 2 formation sigma, close on zero-cross, one-day execution lag, overlapping monthly-start books, committed-capital returns net of engine costs. The claim under test is GGR's: convergence of close twins earns an uncorrelated net return.

Retired at validation: Newey-West t=-0.56 < 3.03 (base 2.5, family n=5); bootstrap P(Sharpe<=0)=0.7414 > 0.01 (base 0.05, family n=5); margin-of-safety rerun (gross halved, costs doubled) t=-1.78 < 2.5, edge has no buffer; in-sample Sharpe …

Tested 2026-08-11

Vidyamurthy common trends, one factor deep. Tests whether purifying the spread rescues the signal or whether the information is simply not there.

Market-residual pairs: matching twins on beta-hedged price paths (formation-window betas to the equal-weight market) removes fake twins whose divergence is a beta gap rather than a relative-value dislocation, and the residual divergence signal beats raw distance divergence.

Retired at validation: Newey-West t=0.42 < 2.87 (base 2.5, family n=3); bootstrap P(Sharpe<=0)=0.3825 > 0.01667 (base 0.05, family n=3); margin-of-safety rerun (gross halved, costs doubled) t=-1.24 < 2.5, edge has no buffer; IS/OOS …

Sharpe 0.087max DD -15%$100k → $105,832 over 20y

Tested 2026-08-11

The genuinely new look in the family: the one-day-delay result inverted into a selection filter. If this fails too, the daily-bar pairs lane closes with receipts.

Slow-spread pairs: restricting the divergence signal to pairs whose formation spread mean-reverts with a half-life of 10 to 60 trading days preserves net-of-lag performance that fast spreads lose to execution delay. The mechanism is latency selection: a spread converging over weeks cannot be destroyed by a one-day-late fill.

Retired at validation: Newey-West t=0.47 < 2.74 (base 2.5, family n=2); bootstrap P(Sharpe<=0)=0.3692 > 0.025 (base 0.05, family n=2); margin-of-safety rerun (gross halved, costs doubled) t=-0.65 < 2.5, edge has no buffer; IS/OOS degradation …

Sharpe 0.104max DD -25%$100k → $109,181 over 20y

Tested 2026-08-11

Baseline and control for the family, on large-cap survivors. Survivorship bias inflates convergence for pairs specifically, both twins survived by construction, so any pass here is an upper bound and must say so.

Distance-pairs divergence as a cross-sectional signal: stocks trading below their formation twin outperform stocks above it (GGR 2006 rules, min-SSD twin over 252d, monthly refresh). Family control ticket: GGR's own one-day-waiting check and a 2026 full-market replication (top-5 CAGR 7.3% to 1.8% when execution waits a day) predict this dies at our built-in 1-day lag.

Retired at validation: Newey-West t=0.22 < 2.50 (base 2.5, family n=1); bootstrap P(Sharpe<=0)=0.4616 > 0.05 (base 0.05, family n=1); margin-of-safety rerun (gross halved, costs doubled) t=-1.23 < 2.5, edge has no buffer; IS/OOS degradation …

Sharpe 0.047max DD -19%$100k → $102,208 over 20y

Tested 2026-08-03

Asks whether how crowded the futures crowd is tells you anything about what equities do next.

S&P 500 futures large-speculator net positioning (CFTC COT), lagged to its public release date, conditions forward returns of the equity panel: positioning extremes precede weaker or more volatile forward returns than neutral positioning

Retired at validation: Not significant, and the return half of the claim points the WRONG WAY. L1_net_length_extreme NW t=-1.24; L2_1y_zscore NW t=-0.76; L3_wow_change NW t=-0.20, all against a deflated bar of 2.87 (base 2.5, family n=3 from …

Tested 2026-07-29

Asks a 25-million-parameter pretrained model to out-forecast a 20-parameter regression on volatility.

Kronos-small (24.7M-param K-line foundation model, zero-shot) beats expanding-window log-Parkinson HAR on 5-day realized volatility (prereg docs/kronos_vol_prereg.md, locked 2026-07-21 before any run)

Retired at the out-of-sample test against a linear benchmark: Pre-registered head-to-head: HAR beats zero-shot Kronos on all 21 tickers (19/21 individually significant), pooled QLIKE 0.148 vs 0.591 (~4x), DM NW-t=-12.37. Verdict existed only in docs since 2026-07-21; filed to the …

Tested 2026-07-28

Tests whether stocks with chunky price grids relative to their volatility trend differently from fine-grid names.

Volatility-normalised tick size conditions short-horizon trend/reversal in US large caps (Kurth-Eisler-Rej-Bouchaud 2026 transfer): high tick-to-dollar-vol names behave like the papers large-tick tier (continuation survives), low ratio like small-tick (dead or reverting)

Retired at validation: none of the 4 declared looks clears the deflated bar (best NW t=0.10 vs t_bar=2.96, family n=4); the conditioning contrast (continuation high-minus-low, ann=6.14%, NW t=3.21) clears the deflated t bar but fails: …

Tested 2026-07-28

One of the oldest anomalies says companies that grow assets fastest underperform - empire-building destroys value. The AI buildout is history's biggest capex boom since WWII (Bridgewater). Is the old rule holding, or is this the cycle that breaks it?

Asset-growth anomaly vs the AI capex boom: the classic high-capex-underperforms effect (Cooper-Gulen-Schill) weakens or inverts for AI-buildout leaders post-2023

Retired at validation: neither declared look clears the joint family gate (n=2, bar t>=2.74): full-sample anomaly D1-D10 NW t=-0.71 (stressed -0.6), post-2023 inversion D10-D1 NW t=0.91 (stressed 0.33)

Tested 2026-07-28

Tests whether the recent gains of a stock's connected peers predict its own next move, beyond its own trend.

Peer-momentum spillover: sector/rolling-correlation-graph peer signals add return information beyond own momentum AND Moskowitz-Grinblatt industry momentum (point-in-time edges, turnover-costed)

Retired at validation: Neither graph's peer-momentum signal carries information beyond own momentum and Moskowitz-Grinblatt industry momentum once those are residualized out cross-sectionally. Sector leave-one-out peer signal: NW t=0.27 vs …

Tested 2026-07-28

Tests whether the market's price of fear and the bond term premium predict WHEN the classic factor strategies pay, instead of assuming they always pay the same.

Time-varying price of risk (Adrian-Crump-Moench JFE 2015 lineage): lagged price-of-risk states (VIX level, ACM term premium, term spread) forecast the FF factor premia time-series, factor timing via priced states, not constant lambdas

Retired at validation: Newey-West t=-0.14 < 2.50 (base 2.5, family n=1); bootstrap P(Sharpe<=0)=0.5682 > 0.05 (base 0.05, family n=1); margin-of-safety rerun (gross halved, costs doubled) t=-0.07 < 2.5, edge has no buffer; IS/OOS degradation …

Tested 2026-07-28

Tests whether American fear prices the whole world's stock markets, and whether emerging markets load on it harder than developed ones.

Global price of risk (Adrian-Stackman-Vogt IMF ER 2019): the US-VIX-based price of risk forecasts international equity returns (EFA/EEM) with heterogeneous loadings, US fear is a global state variable

Retired at validation: No leg clears the gates (family_n=3, deflated bar t>=2.87). EFA leg is dead (NW t=0.11). EEM leg: NW t=2.19 < 2.87, stressed t=1.09, IS/OOS degradation 37%. The heterogeneity leg (EEM-EFA, the paper's core claim) clears …

Tested 2026-07-28

The BIS's 2026 warning in one sentence: hyperscalers flipped from cash machines to net borrowers to fund AI. The oldest financing anomaly says heavy issuers underperform. Is the oldest rule holding through the loudest capex boom - or is this the exception the borrowers are betting on?

External-financing anomaly, AI-capex edition: heavy net debt issuers among AI-buildout names underperform (BIS 2026: top-5 AI commitments now exceed free cash flow; the classic Bradshaw-Richardson-Sloan effect meets the biggest debt-funded capex boom since the telecoms)

Retired at validation: Right sign, no significance. Trailing-12m BRS net external financing (PIT from EDGAR companyfacts cash-flow XBRL, 435 tickers, 27,772 PIT events, coverage from 2010-05), quintile L/S long retirers / short heavy issuers: …

Tested 2026-07-28

A weather cycle with a 75-year public index and a mechanical pass-through story (crops, cooling demand, hydro shortfalls). Kin to the retired sunshine study but a different animal: seasonal-scale supply shocks, not daily mood. CIBC's own rule of thumb - normal El Nino adds ~5% to commodity inflation - is the claim I make testable in equities.

El Nino regimes (NOAA ONI index) predict relative returns of commodity-sensitive equity sectors: energy and staples names outperform the panel during El Nino phases as food and fuel inflation passes through

Retired at validation: Newey-West t=0.24 < 2.50 (base 2.5, family n=1); bootstrap P(Sharpe<=0)=0.7298 > 0.05 (base 0.05, family n=1); margin-of-safety rerun (gross halved, costs doubled) t=0.12 < 2.5, edge has no buffer; IS/OOS degradation …

Tested 2026-07-28

A global affordability map (UN Habitat via Visual Capitalist) shows the US at 4.5x income - 7th most affordable of 181 - while China sits at 34.6x. The testable US question: does a stretched domestic affordability ratio predict homebuilder underperformance? Intuitive, probably weak - a tested rejection is the likely honest outcome, and that is museum content.

Housing affordability ceiling: when the US home-price-to-income ratio is stretched (top tercile), US homebuilder stocks (DHI/LEN/PHM/NVR) subsequently underperform the panel over the next 1-2 quarters as affordability caps demand

Retired at validation: Newey-West t=0.67 < 2.50 (base 2.5, family n=1); bootstrap P(Sharpe<=0)=0.1873 > 0.05 (base 0.05, family n=1); margin-of-safety rerun (gross halved, costs doubled) t=0.33 < 2.5, edge has no buffer; IS/OOS degradation …

Tested 2026-07-28

Table 8.1 row 3: income funds forced to sell on a dividend cut may push price below fair value, creating a reversal. Testable event study; needs a dividend-action date harvest.

Dividend-cut forced-selling overshoot (Manual of Ideas Table 8.1, 'dividend cancellation'): stocks cutting/suspending dividends overshoot to the downside on yield-seeker forced selling, then partially reverse over the next 1-3 months

Retired at validation: reversal (tradeable) leg fails the gates: margin-of-safety rerun (gross halved, costs doubled) t=1.85 < 2.5, edge has no buffer

Tested 2026-07-26

Kalshi calibration, favourite-longshot bias, and crowding

Prediction markets (Kalshi) are calibrated, and one-sided books resolve against the crowd (pre-registered: docs/prediction_market_prereg.md, family n=3)

Retired at pipeline: 250,858 resolved Kalshi markets across 59,748 distinct events, price = VWAP over the final 24h. C1 CALIBRATION FALSIFIED: the reliability curve departs sharply from the diagonal (reliability component 0.019 of a 0.093 …

Tested 2026-07-26

Tests whether trading crowding into a few giant stocks foretells weaker stock-picking signals and lagging small stocks.

Concentration regime (top-10 share of dollar volume, rolling) conditions the EW-vs-CW spread and cross-sectional signal ICs, rising concentration predicts EW underperformance and IC compression (DESCO Concentration Game 2026)

Retired at pipeline: Answered under the concentration_lab study: EW lags CW -3.3%/yr when the market is top-heavy vs -0.75% when broad, but the regime-timed EW/CW tilt fails (t=-0.68). Duplicate; see the concentration tombstone.

Tested 2026-07-26

Tests whether gold's value in a stock-bond portfolio depends on whether stocks and bonds move together or apart.

Gold conditional diversification: stock-bond correlation regime (rolling SPY/TLT) conditions gold allocation value, 60/40 vs 60/35/5+GLD by SB-corr sign (DESCO Worth Its Weight)

Retired at pipeline: Answered under the gold_lab study: gold is a real conditional diversifier (19.3%/yr in positive stock-bond-correlation regimes vs 9.8%) but the regime-timed overlay fails the gates (t=-1.04). Duplicate; see the gold …

Tested 2026-07-26

Tests whether dropping energy stocks from a portfolio secretly amounts to a bet on cheap versus expensive stocks.

Sector-exclusion tilts are factor bets in disguise: ex-energy active returns largely explained by HML loading (DESCO Factored In)

Retired at pipeline: Eleven sector exclusions tested on 21 years. Excluding a sector moves the active return by under 1%/yr in every case and only two are even nominally significant (ex-tech -0.73%/yr t=-2.56, ex-consumer-discretionary …

Tested 2026-07-26

MIT scored S&P 500 AI adoption from SEC filings; a replicator found the return premium vanishes once the 38 biggest companies leave the sample - the AI premium may be mega-cap beta in costume. I test adoption with the confounder control built in from day one.

AI-adoption language in 10-K filings carries residual alpha after size and factor control (MIT scoring, Increase Alpha's mega-cap critique applied from the start)

Retired at pipeline: 14,155 EDGAR filings across 406 names scored for AI vocabulary per 10,000 words. AI language exploded 76x (0.01 in 2015 to 0.76 in 2026), but it does NOT predict returns: Fama-MacBeth t=1.80 controlling for momentum, …

Tested 2026-07-26

cross-sectional panel regression, all characteristics at once

Fama-MacBeth cross-sectional panel: which characteristics are priced in US large caps once momentum, reversal, volatility, size, illiquidity and short-volume positioning are held constant

Retired at pipeline: Fama-MacBeth on 5,170 dates (348 names/date). Controlling for each other, momentum is NOT priced (t=1.22) and illiquidity is not either over the full history (t=1.26). Short-term reversal is the strongest price …

Tested 2026-07-26

EDGAR filing-language change (forensic text, orthogonal)

Disclosure text-change predicts returns (Cohen-Malloy-Nguyen 'Lazy Prices'): when a firm rewrites its 10-Q/10-K vs the prior year, its stock underperforms

Retired at pipeline: 14,155 EDGAR 10-Q/10-K filings, 406 names. A year-over-year cosine text-change measure is methodology-sensitive and does NOT replicate Lazy Prices here: cleaned and seasonally controlled it is significant but …

Tested 2026-07-26

market concentration / breadth regime (topical macro)

Regime-timed equal-weight/cap-weight tilt: lean to equal-weight when market concentration is falling, cap-weight when rising

Retired at pipeline: Concentration is a real descriptive regime, EW lags CW -3.3%/yr when top-heavy vs -0.75%/yr when broad, but the regime-timed EW/CW tilt earns -1.28%/yr at NW t=-0.68 (bar 2.5), fails. A real regime, not a tradable …

Tested 2026-07-26

gold as a conditional diversifier (macro/regime)

Regime-timed gold overlay: hold gold when the stock-bond correlation is positive (bonds stop hedging), a tradable macro-diversification signal

Retired at pipeline: Gold returned more in the positive stock-bond-correlation regime (19.3%/yr vs 9.8%), but its 60/40 vol-reduction is regime-invariant (~0.4-0.6%, slightly better when bonds hedge), and the regime-timed overlay earns …

Tested 2026-07-26

FINRA short-volume positioning (free, orthogonal)

Short-side crowding (FINRA daily short-volume ratio) predicts large-cap cross-sectional returns, and flags stress underperformance

Retired at pipeline: Free FINRA short-volume panel (2006 days x 406 names, 2017-12-29..2026-07-24). Family of 4 cross-sectional SVR signals all fail the deflated bar t>=2.96 (best |t|=2.24). Conditional risk measure is loud, most-shorted …

Tested 2026-07-26

Altucher calendar/flow family (non-price market-timing)

End-of-quarter window-dressing effect (Altucher #12): returns around quarter turns, equal-weight universe

Retired at pipeline: Quarter-end spread +5.87 bps/day, gross +0.89%/yr, NW t=0.83, not significant against the deflated 3.03 bar. Tested and shelved.

Tested 2026-07-26

Altucher calendar/flow family (non-price market-timing)

Option-expiration-week effect (Altucher #14): the five days ending on the 3rd-Friday expiry, a dealer-positioning calendar

Retired at pipeline: Expiry week underperforms mildly (4.62 bps/day, gross 2.09%/yr) at only NW t=-1.19; no gate-clearing edge either direction. Tested and shelved.

Tested 2026-07-26

Altucher calendar/flow family (non-price market-timing)

Weekend / Monday effect (Altucher day-of-week family): Mondays vs the rest of the week, equal-weight universe

Retired at pipeline: Weekend effect has decayed: Mondays underperform by ~6.62 bps/day but only at NW t=-1.35 gross, it reverses post-2013 (subperiod t 0.99->1.19->1.29->-0.22), and twice-weekly rebalancing costs (~4.71%/yr) swamp the …

Tested 2026-07-26

Altucher calendar/flow family (non-price market-timing)

Pre-holiday effect (Altucher / Ariel 1990): last trading day before a market holiday, equal-weight universe

Retired at pipeline: Pre-holiday days average +14.82 bps vs the rest, a real magnitude, but only ~3% of days: gross +1.33%/yr at NW t=2.11, below even the naive 2.5 bar and well under the family-deflated 3.03. Underpowered, tested and …

Tested 2026-07-25

A public library of 158 alphas that half the industry has mined is the perfect multiplicity teaching exhibit: with the family honestly declared at n=158, the deflator does to a famous alpha library what it did to my own founder's model. Museum series material.

The Alpha158 excavation: Microsoft Qlib's 158 public technical alphas, run as ONE family through the museum's gates - how many famous library alphas survive honest deflation?

Retired at pipeline: Alpha158 family (158 alphas): 29 clear naive t>=2.5, 0 are promoted. The strongest member, IMAX20, DOES clear the deflated bar (t=4.46 >= 4.11) and the margin-of-safety rerun (stressed t=3.68 >= 2.50); it is stopped by …

Tested 2026-07-24

Distinct from my rejected TLH-cost study: this is the OPPORTUNITY side - buying the victims of December tax-loss selling for the January bounce. Table 8.1's first row, made testable.

Tax-loss-selling bounce (Manual of Ideas Table 8.1, 'end-of-year tax selling'): stocks most depressed into year-end (biggest YTD losers) outperform in January as tax-loss selling pressure lifts

Retired at validation: January tax-loss bounce: worst-decile losers beat the panel by +179bps over Dec20-Jan31, NW t=2.20 on 20 years - the classic January effect does not clear the museum bar in the recent large-cap era (survivorship-biased …

Tested 2026-07-24

Fama-French declared HML redundant in the US in 2015 (subsumed by profitability and investment). A 2026 deep-research sweep found HML is NOT redundant internationally. Simplest honest question: is it redundant in OUR specific US large-cap, survivorship-biased, recent-era panel? A clean replication with a public verdict either way.

HML redundancy on home turf: is the value factor subsumed by RMW+CMA+Mom in my US large-cap panel 2005-2026, as Fama-French (2015) claimed for the US - or does it retain alpha, as post-2015 international evidence finds?

Retired at validation: no promotable factor alpha: HML redundant in US large-cap (alpha -0.4%/yr full, +0.9%/yr post-2015, |t|<0.3 both) - a clean replication of Fama-French 2015's US redundancy result; CMA also redundant (t=0.25); only RMW …

Tested 2026-07-24

My regime auditor is an HMM. A Princeton ORFE paper claims the statistical jump model - k-means through time with a penalty for switching - is more robust and persistent. Incumbent vs challenger, same allocation rule, same data, out-of-sample. If the challenger wins, the museum's own stage 5 goes under review; if it loses, the HMM earns its seat in public.

Statistical jump model as stage-5 challenger (Shu-Yu-Mulvey 2024): temporal k-means with an explicit jump penalty identifies more persistent regimes than my HMM - a regime-aware SPY/cash rule on SJM signals beats the SAME rule on HMM signals out-of-sample

Retired at validation: challenger fails to displace the incumbent: SJM-rule vs HMM-rule monthly diff +17.5bps/mo, NW t=0.68 (statistical noise; gates need 2.5 and stressed 2.5). On the 2017-2026 SPY testbed the incumbent HMM delivered the …

Tested 2026-07-24

The oldest 'free lunch' in indexing - buy the stock entering the S&P before the trackers must - reportedly shrank to nothing (Greenwood-Sammon). Yet the largest platforms' index-rebal teams are cited as top earners. Either the effect is dead and the desks earn something subtler, or the death is exaggerated. A verdict either way is museum-grade.

The index effect is dead, officially: S&P 500 additions no longer earn announcement-to-inclusion abnormal returns (Greenwood-Sammon), yet index-rebalancing desks reportedly print, test where the residual edge migrated

Retired at validation: index effect confirmed dead in the panel era: post-effective CARs are statistical zero in all three eras (2005-14: -97bps t=-0.8; 2015-19: +1bp t=0.0; 2020+: -35bps t=-0.2); the only above-Bonferroni window (+170bps …

Tested 2026-07-23

Bets that stocks owned by the same many hedge funds get hit hardest when those funds all rush the exit at once.

Crowded-exit risk: stocks with high hedge-fund ownership overlap (13F-derived, point-in-time by filing date) underperform specifically during vol-spike windows, the pod-liquidation cascade priced as a conditional factor (Young&Calculated pod-contagion mechanism; Patterson/DESCO lenses as priors)

Retired at validation: comovement diff +0.009 (t=4.09), stress spread -142bps (t=-3.90, n=14), neither claim clears the deflated stressed bar

Tested 2026-07-23

Cliff Asness's argument (via Busca podcast): weighting by inverse volatility is the benchmark no optimizer reliably beats. Equal-weight 7Twelve just died; this asks whether risk-weighting, not asset count, was the missing ingredient.

Asness equal-risk benchmark: inverse-volatility weighting of the 7Twelve 12-ETF sleeve beats its equal-weight version and 60/40 SPY/AGG

Retired at validation: inverse-vol 7Twelve: -4.00%/yr vs equal-weight (t=-1.63), -6.59%/yr vs 60/40 (t=-3.58), risk-weighting was not the missing ingredient either

Tested 2026-07-21

Watches Americans falling behind on credit cards and bets the lenders' stocks feel it after the data drops.

Consumer-credit deterioration predicts consumer-finance equity underperformance: rising card delinquency rates (FRED DRCCLACBS, release-date aligned) lead relative returns of consumer lenders vs market (KKR JD stress-scenario translation)

Retired at validation: rising-vs-falling delinquency: -1.2pct/qtr relative, t=-0.51 on 87 quarters, declared-underpowered test, honest inconclusive

Tested 2026-07-21

Splits money equally across twelve funds in seven asset classes, betting that true breadth lives across asset classes, not across more stocks.

7Twelve multi-asset equal-weight (Israelsen 2010) beats 60/40 and SPY on geometric growth and drawdown via cross-asset-class breadth

Retired at validation: 7Twelve vs 60/40: CAGR diff -2.51%/yr, t=-1.82, cross-asset equal weight did not beat the boring benchmark on this sample

Tested 2026-07-21

Tests the famous claim that traders buy more on sunny mornings in New York, mood as a market force.

Sunshine effect replication (Hirshleifer-Shumway JF 2003): morning cloud cover in New York predicts daily index returns, the classic behavioral claim, retested 2005-2026 under modern gates

Retired at validation: gloom-day effect -1.2bps/day, t=-0.34 (PRCP>0 (precipitation day)), the famous mood effect does not clear modern gates on 2005-2026 data

Tested 2026-07-13

Tests whether scary markets pay stock holders extra while bond prices have already flown to safety, fear as a priced premium, not just a mood.

Flight-to-safety nonlinearity (Adrian-Crump-Vogt JF 2019): 1m forward stock returns RISE with lagged VIX while Treasury returns FALL, replication on SPY/TLT 2012-2026 with post-publication subsample

Retired at validation: Directionally replicates ACV in AND out of sample, monotone SPY rise, TLT mirror, but t=2.25 sits under the 2.5 bar with overlapping-window inference, and far under the 5.0 stressed promotion bar. Verdict: the paper's …

Tested 2026-07-13

Holds more stock when markets are calm and less when they are stormy, betting risk isn't paid extra in high-vol months.

Vol-managed SPY (Moreira-Muir 2017): scale exposure by inverse realized variance, cap 2x, higher Sharpe than buy-and-hold

Retired at validation: gate t=1.55; promotion needs t>=5.0 (2.5 base + margin-of-safety)

Tested 2026-07-13

Buys boring low-beta stocks and shorts exciting high-beta ones, levering the boring side to match risk.

Betting-against-beta (Frazzini-Pedersen 2014): leverage-constrained investors overpay for high beta; L/S beta-neutral spread earns a premium on my panel

Retired at validation: gate t=0.29; promotion needs t>=5.0 (2.5 base + margin-of-safety)

Tested 2026-07-13

Owns stocks only around month-end, betting paycheck and fund flows push prices up on a calendar schedule.

Turn-of-month effect (McConnell-Xu 2008): SPY returns concentrate in the last 2 + first 3 trading days of the month (mechanical month-end flows)

Retired at validation: gate t=0.64; promotion needs t>=5.0 (2.5 base + margin-of-safety)

Tested 2026-07-13

Adds a small trend-following sleeve to a stock portfolio, betting the hedge cheapens crashes enough to speed up long-run compounding.

A 10-20% time-series-momentum sleeve improves the GEOMETRIC growth and max drawdown of an equity book vs equity-only (crisis-alpha portfolio construction; Moskowitz-Ooi-Pedersen lineage). Distinct from the rejected standalone walk-forward TSMOM family: this claims drag reduction at the PORTFOLIO level, not timing alpha

Retired at validation: 80/20 trend sleeve: CAGR diff -0.06%/yr, t=-0.10 vs bar 2.74 + stressed 2.5, did not clear

Tested 2026-07-13

Holds the index as individual stocks and systematically sells losers to bank tax losses, swapping into similar names to keep tracking the index.

Tax-loss harvesting adds ~1%/yr after-tax alpha in early years on a direct-indexed S&P portfolio, decaying as embedded gains age (Chaudhuri-Burnham-Lo 2020 replication; the Parametric mechanism)

Retired at validation: full-liquidation tax alpha -1.75%/yr, t=-2.34: the pre-tax bleed of sell-the-losers swaps (~180bps/yr) dwarfs the tax credits (~6bps/yr on a no-contribution book). Two honest caveats: (1) survivorship interaction, on a …

Tested 2026-07-13

Tests whether free market-fear gauges forecast worst-case daily stock losses better than standard statistical models.

Quantile-regression VaR on SPY from free implied-moment proxies (VIX level, VIX/VIX3M term slope, CBOE SKEW) beats GARCH-based and historical-simulation VaR on coverage and pinball loss (Blom/de Lange/Risstad JRFM 2023, EURUSD -> US equity translation)

Retired at validation: primary cell: FAIL

Tested 2026-07-13

Tests whether knowing option dealers' positioning sharpens forecasts of the market's worst days, not its typical ones.

GEX states condition the TAILS, not the mean: adding dealer-positioning state to the implied-moment quantile regression improves tail coverage where the Chain's mean forecast failed (chain-family lineage declared for the deflator; per-state residue s2/s6 from v3 is the motivating anomaly, mechanism = short-gamma amplification is a tail phenomenon)

Retired at validation: primary cell: FAIL

Tested 2026-07-13

Tests whether option-dealer positioning and market state improve five-day forecasts of how far the S&P 500 will swing.

Dimopoulos Chain v3: shrunk state corrections (GEX x RV cells + term structure + arrival) on a log HAR+VIX+GJR backbone improve 5-day range forecasts (pre-registered spec)

Retired at validation: FAIL, buried, name and all

Tested 2026-07-13

Tests whether dealer hedging plus recent choppiness predicts tomorrow's trading range beyond the market's fear gauge.

Dimopoulos Chain v1: GEX x RV state chain adds next-day range information beyond VIX (OOS, monthly-refit)

Retired at validation: margin-of-safety rerun (edge halved vs VIX baseline, original noise) t=2.17 < 2.5, edge has no buffer; gate tightened 2026-07-12 (Graham overlay), prior pass was under the old regime

Tested 2026-07-09

Trains a neural network to imitate a slow options-pricing model, trading a long calculation for an instant lookup.

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

Retired at a later revision that superseded it: CORRECTED 2026-07-29 (audit of vault against published notes, per lessons/rerun-verdicts-must-overwrite.md rule 2). The original reason recorded here, 'MAE 66.86 vp / speedup 51684x missed gates', was a BUG ARTIFACT, …

Tested 2026-07-09

Trains a small neural network on hedging and volatility signals to forecast the size of tomorrow's market moves.

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

Retired at the out-of-sample test against a linear benchmark: neural emissions did not beat linear ladder (t=-4.37)

Tested 2026-07-09

Bets on market calm but steps aside whenever warning signs point to a sudden violent snap-back in fear.

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

Retired at the excess test against always being long: no significant excess over always-long (t=-0.36, p=0.660)

Tested 2026-07-09

Bets on market calm, but only when insurance against swings costs more for later dates than for the near term.

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

Retired at the excess test against always being long: no significant excess over always-long (t=0.57, p=0.313)

Tested 2026-07-09

Bets on market calm only when swing insurance is pricier further out and dealers' hedging is steadying the market.

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

Retired at the excess test against always being long: no significant excess over always-long (t=-1.72, p=0.962)

Tested 2026-07-09

Bets on market calm only on days when options dealers' hedging tends to pin the market in place.

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

Retired at the excess test against always being long: no significant excess over always-long (t=-1.68, p=0.959)

Tested 2026-07-09

Holds more stock when options dealers' hedging tends to calm the market, and less when it tends to amplify swings.

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

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

Tested 2026-07-08

Tests whether dealer-hedging and choppiness states improve on a full battery of standard volatility forecasters.

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 the out-of-sample test against the full baseline: 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 …

Tested 2026-07-08

Follows the stock market's trend, re-picking the best-performing trend rule from recent years as it goes.

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

Retired at the out-of-sample excess test: stitched OOS excess vs B&H: t=-0.62, p=0.738

Tested 2026-07-08

Tests whether riding the stock market's trend, and easing off when it turns choppy, beats simply holding it.

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

Buys companies now reporting longer lifespans for their equipment, betting built-to-last firms are underpriced.

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 the BRAIN screen: LOW_SHARPE, LOW_FITNESS, CONCENTRATED_WEIGHT

Tested 2026-07-08

Tests whether sudden bursts of local news coverage about small companies come before unusual stock moves.

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

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

Tested 2026-07-07

Buys hard-to-trade large stocks whose prices swing on light trading, betting investors earn extra for the hassle.

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

Buys last month's losers and sells its winners, this time only among the 500 largest US stocks.

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

Buys the past year's winners, skipping the latest month, and sells its losers, only among the 500 largest US stocks.

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

Buys large stocks trading far more heavily this week than usual, betting the burst of attention pulls in new buyers.

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

Buys large stocks whose surprises tilt toward drops rather than jumps, betting jackpot-style shares are overpriced.

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

Buys large stocks with no big one-day pop in the past month, betting lottery-style shares are overpriced.

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

Buys large stocks with the least stock-specific wobble, betting investors overpay for excitement.

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

Buys stocks whose gains come on heavy trading over three months, betting trends confirmed by volume keep going.

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

Buys stocks sitting near their highest price of the past year, betting hesitant investors let winners keep climbing.

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

Buys last month's losers and sells its winners, betting crowds overreact and prices snap back.

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

Buys the past year's biggest gainers, skipping the most recent month, and sells the laggards, betting trends persist.

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

Every idea runs the same six gates: 1 idea sourcing · 2 point-in-time features · 3 backtest · 4 validation (Newey–West, 10,000-draw bootstrap, in/out-of-sample decay) · 5 regime audit · 6 six-factor decomposition. Verdicts publish on a 3-day lag; strategies that pass all six are not published. They go to the vault. Rejection here means the claim failed as I 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 at collection.json. Study it, remix it, prove me wrong. (After the Rijksmuseum’s Rijksstudio.)

Membership opens with the first exhibits.

A retiring athlete put it best, in a farewell written as a museum exhibit: the greatest thing left behind does not fit inside the museum. Same here. The exhibits are the receipts. What built them does not hang on a wall.

Tested Strategies

I rebuild the data layer when the vendor’s is wrong

I pin every input, then try to break the pin

My tests refuse to publish

Selected work

The published factor zoo does not survive an honest bar

I test my own hypotheses to destruction, then audit the test

Ten desks behave like three

Forecasts priced in advance, scored in public

What this record is meant to show

Selected working papers

Stocks that double in two years go on to crash more often. These just did.

Tested strategies: the permanent collection

204 machine-written alpha expressions, one collective tombstone

13F shadow of equity held on swap, via dealer entities with no asset-management arm

The uncontaminated version of the slow-spread look. Judged after the headline variant, n=5 at judge time.

The critic-prescribed fair test after three cross-sectional conversions died. Survivorship inflates convergence for pairs specifically, so any pass is an upper bound. Judged at the family-deflated bar, n=4 at judge time.

Vidyamurthy common trends, one factor deep. Tests whether purifying the spread rescues the signal or whether the information is simply not there.

The genuinely new look in the family: the one-day-delay result inverted into a selection filter. If this fails too, the daily-bar pairs lane closes with receipts.

Baseline and control for the family, on large-cap survivors. Survivorship bias inflates convergence for pairs specifically, both twins survived by construction, so any pass here is an upper bound and must say so.

Asks whether how crowded the futures crowd is tells you anything about what equities do next.

Asks a 25-million-parameter pretrained model to out-forecast a 20-parameter regression on volatility.

Tests whether stocks with chunky price grids relative to their volatility trend differently from fine-grid names.

One of the oldest anomalies says companies that grow assets fastest underperform - empire-building destroys value. The AI buildout is history's biggest capex boom since WWII (Bridgewater). Is the old rule holding, or is this the cycle that breaks it?

Tests whether the recent gains of a stock's connected peers predict its own next move, beyond its own trend.

Tests whether the market's price of fear and the bond term premium predict WHEN the classic factor strategies pay, instead of assuming they always pay the same.

Tests whether American fear prices the whole world's stock markets, and whether emerging markets load on it harder than developed ones.

The BIS's 2026 warning in one sentence: hyperscalers flipped from cash machines to net borrowers to fund AI. The oldest financing anomaly says heavy issuers underperform. Is the oldest rule holding through the loudest capex boom - or is this the exception the borrowers are betting on?

Table 8.1 row 3: income funds forced to sell on a dividend cut may push price below fair value, creating a reversal. Testable event study; needs a dividend-action date harvest.

Kalshi calibration, favourite-longshot bias, and crowding

Tests whether trading crowding into a few giant stocks foretells weaker stock-picking signals and lagging small stocks.

Tests whether gold's value in a stock-bond portfolio depends on whether stocks and bonds move together or apart.

Tests whether dropping energy stocks from a portfolio secretly amounts to a bet on cheap versus expensive stocks.

MIT scored S&P 500 AI adoption from SEC filings; a replicator found the return premium vanishes once the 38 biggest companies leave the sample - the AI premium may be mega-cap beta in costume. I test adoption with the confounder control built in from day one.

cross-sectional panel regression, all characteristics at once

EDGAR filing-language change (forensic text, orthogonal)

market concentration / breadth regime (topical macro)

gold as a conditional diversifier (macro/regime)

FINRA short-volume positioning (free, orthogonal)

Altucher calendar/flow family (non-price market-timing)

Altucher calendar/flow family (non-price market-timing)

Altucher calendar/flow family (non-price market-timing)

Altucher calendar/flow family (non-price market-timing)

A public library of 158 alphas that half the industry has mined is the perfect multiplicity teaching exhibit: with the family honestly declared at n=158, the deflator does to a famous alpha library what it did to my own founder's model. Museum series material.

Distinct from my rejected TLH-cost study: this is the OPPORTUNITY side - buying the victims of December tax-loss selling for the January bounce. Table 8.1's first row, made testable.

Bets that stocks owned by the same many hedge funds get hit hardest when those funds all rush the exit at once.

Cliff Asness's argument (via Busca podcast): weighting by inverse volatility is the benchmark no optimizer reliably beats. Equal-weight 7Twelve just died; this asks whether risk-weighting, not asset count, was the missing ingredient.

Watches Americans falling behind on credit cards and bets the lenders' stocks feel it after the data drops.

Splits money equally across twelve funds in seven asset classes, betting that true breadth lives across asset classes, not across more stocks.

Tests the famous claim that traders buy more on sunny mornings in New York, mood as a market force.

Tests whether scary markets pay stock holders extra while bond prices have already flown to safety, fear as a priced premium, not just a mood.

Holds more stock when markets are calm and less when they are stormy, betting risk isn't paid extra in high-vol months.

Buys boring low-beta stocks and shorts exciting high-beta ones, levering the boring side to match risk.

Owns stocks only around month-end, betting paycheck and fund flows push prices up on a calendar schedule.

Adds a small trend-following sleeve to a stock portfolio, betting the hedge cheapens crashes enough to speed up long-run compounding.

Holds the index as individual stocks and systematically sells losers to bank tax losses, swapping into similar names to keep tracking the index.

Tests whether free market-fear gauges forecast worst-case daily stock losses better than standard statistical models.

Tests whether knowing option dealers' positioning sharpens forecasts of the market's worst days, not its typical ones.

Tests whether option-dealer positioning and market state improve five-day forecasts of how far the S&P 500 will swing.

Tests whether dealer hedging plus recent choppiness predicts tomorrow's trading range beyond the market's fear gauge.

Trains a neural network to imitate a slow options-pricing model, trading a long calculation for an instant lookup.

Trains a small neural network on hedging and volatility signals to forecast the size of tomorrow's market moves.

Bets on market calm but steps aside whenever warning signs point to a sudden violent snap-back in fear.

Bets on market calm, but only when insurance against swings costs more for later dates than for the near term.

Bets on market calm only when swing insurance is pricier further out and dealers' hedging is steadying the market.

Bets on market calm only on days when options dealers' hedging tends to pin the market in place.

Holds more stock when options dealers' hedging tends to calm the market, and less when it tends to amplify swings.

Tests whether dealer-hedging and choppiness states improve on a full battery of standard volatility forecasters.

Follows the stock market's trend, re-picking the best-performing trend rule from recent years as it goes.

Tests whether riding the stock market's trend, and easing off when it turns choppy, beats simply holding it.

Buys companies now reporting longer lifespans for their equipment, betting built-to-last firms are underpriced.

Tests whether sudden bursts of local news coverage about small companies come before unusual stock moves.

Buys hard-to-trade large stocks whose prices swing on light trading, betting investors earn extra for the hassle.

Buys last month's losers and sells its winners, this time only among the 500 largest US stocks.

Buys the past year's winners, skipping the latest month, and sells its losers, only among the 500 largest US stocks.

Buys large stocks trading far more heavily this week than usual, betting the burst of attention pulls in new buyers.

Buys large stocks whose surprises tilt toward drops rather than jumps, betting jackpot-style shares are overpriced.

Buys large stocks with no big one-day pop in the past month, betting lottery-style shares are overpriced.

Buys large stocks with the least stock-specific wobble, betting investors overpay for excitement.

Buys stocks whose gains come on heavy trading over three months, betting trends confirmed by volume keep going.