THE STRATEGY LIBRARY

In 1965 to 1989 the 6-month/6-month decile strategy earned 12.01% per year compounded (about 0.95% per month); the best variant (12-month formation, 3-month hold, one-week gap) earned about 1.5% per month. Part of the gain reversed in years 2 and 3 and the strategy lost money in Januaries. Profits continued in 1990 to 1998 (Jegadeesh and Titman 2001), with severe crashes in rebounds such as 1932 and 2009 (Daniel and Moskowitz 2016).

Over 1962 to 1993, the hot-hands persistence of top-performing funds was largely momentum exposure plus expenses, not skill; the only significant persistence left was among the worst funds. The paper builds a factor, not a stock strategy; the later value-weighted UMD (Ken French library) is the common momentum benchmark.

In US stocks 1963 to 2001 the strategy earned about 0.45% per month, about 1.23% excluding January, and dominated individual and industry momentum in head-to-head tests. Unlike past-return momentum, its profits did not reverse in later years.

All 58 instruments showed positive time-series momentum over data from as early as 1965 to 2009. The diversified portfolio did best in extreme markets, including 2008, and explains much of managed-futures returns. Returns partially reverse beyond 12 months.

Jegadeesh found a 2.49% per month abnormal return difference between extreme deciles over 1934 to 1987; Lehmann found weekly reversal profits that survived his bid-ask and cost corrections. Later studies find the profits concentrate in small, illiquid stocks and shrink sharply after realistic trading costs.

Over 1977 to 1993, past returns and past earnings surprises each predicted sizeable later drifts after controlling for the other, with no sign of later reversal; market risk, size and book-to-market did not explain them.

Graham offered the screen as a safety standard and published no backtest. Tests of the exact seven rules are scarce, and results depend on how the size, 10-year and 20-year history tests are adapted.

HML averaged 0.40% per month (about 4.8%/yr, t = 2.91) from July 1963 to December 1991. Value then had a long drawdown from about 2007 to 2020, with a partial recovery since.

1976-1996: high F-Score (8-9) value stocks beat the average high-B/M stock by at least 7.5%/yr; long high minus short low (0-1) earned 23%/yr. Gains concentrate in small, thinly traded names with no analyst coverage.

Value beat glamour by about 10 to 11%/yr over the 5 years after formation in their 1968-1990 sample, and did not do worse in down markets or recessions.

In a study Carlisle commissioned (1973-2017, 30 stocks), it beat the Magic Formula by about 0.3 to 2.4 points a year depending on the cap floor, and both beat the S&P 500. Author-run tests; long losing stretches occur.

Trending Value compounded about 21%/yr from 1964 to 2009 in the 4th edition's backtest (before costs), well ahead of the All Stocks universe, with deep drawdowns along the way.

Buffett (1984): 16.1%/yr to limited partners vs 8.4% for the S&P, 1956-1984 (21.3%/yr before Schloss's share). Commonly cited 47-year record: about 15.3%/yr net vs about 10% for the S&P. A discretionary record, not a tested rule.

April 1957 to March 1971: low-P/E portfolios earned higher raw and risk-adjusted returns than high-P/E portfolios, with returns falling roughly in order of P/E.

Stattman reported that US stocks with higher book-to-market earned higher average returns; Fama and French (1992) cite it with Rosenberg, Reid and Lanstein (1985) as early evidence of the value effect.

Dreman compiled evidence that unpopular low-P/E stocks beat glamour stocks over the periods he studied; he stresses that single holdings can be volatile. No single backtest figure is given here.

An enterprise-multiple factor built Fama-French style earned a 5.28%/yr premium, July 1963 to December 2009; low-EM firms had higher later returns.

Gross profits-to-assets predicted returns about as well as book-to-market, and profitable firms outperformed despite higher valuations. Controlling for profitability sharply improved value strategies, especially among the largest, most liquid stocks.

Book backtest 1974-2011: 17.68%/yr vs 10.46% for the S&P 500, about 3.7 points a year ahead of the Magic Formula. Author-run and before costs.

1963-2010: top minus bottom quintile earned 0.31%/month (t = 2.49), FF3 alpha 0.52%/month. It lagged from the mid-1970s to the early 1980s and in the mid-2000s.

1962-1991: the hedge earned about 10.4% size-adjusted in year one, positive in 28 of 30 years, smaller after. Later studies (Green, Hand and Soliman 2011) find it largely faded in the 2000s.

Significant risk-adjusted returns in the US (1957-2016) and across 24 countries. QMJ has negative market beta and tended to do well in sell-offs.

1968-2003: low minus high decile about 8%/yr value weighted, much larger equal weighted, and it held in large caps. Fama-French (2015) built CMA on asset growth.

Practitioner backtests (Faber 2013) favor it over dividend yield. Boudoukh et al. (2007) found net payout yield predicts returns better than dividend yield.

A practitioner style, not a published tested strategy. Mauboussin's data show high ROIC tends to fade toward the average, so persistence has to be checked, not assumed.

About 1971-2000: lowest minus highest financing decile earned about 15.5% a year size-adjusted, and heavy issuers drew over-optimistic analyst forecasts. Cohen and Lys (2006) find it weakens once accruals are controlled.

Titman, Wei, and Xie report negative benchmark-adjusted returns after large investment increases, stronger for firms with more cash and less debt.

Earnings sign lined up with abnormal returns, but about 85 to 90 percent of the information was priced before the announcement month. Post-announcement drift was a small side finding, later formalized by Bernard and Thomas (1989).

Extreme-decile spread of about 4.2% size-adjusted over 60 trading days (about 18% annualized before costs), NYSE/AMEX 1974 to 1986, positive in 41 of 48 quarters and larger in small firms. Later studies report it has weakened, especially in large caps.

12.1% average abnormal buy-and-hold return over four years for 1980 to 1990 announcements, and 45.3% for value stocks. Fu and Huang (2016) find the long-run drift disappeared for 2003 to 2012 announcements.

O'Neil describes the cup with handle as one of the most common bases among past big winners; that comes from winners-only chart studies with no failure rate, so breakout results must be tested.

Minervini won the US Investing Championship in 1997 (155%) and 2021 (334.8%, $1 million division) with his full discretionary SEPA trading; those results are not a test of this screen. The template itself is mechanical and testable.

By Darvas's own unaudited account, about $36,000 became about $2.25 million in roughly 18 months in the late-1950s bull market, from a handful of stocks. One trader's record, not a tested system.

Livermore's own rules, written late in his career; they are partly discretionary (Market Key records, timing) and come with no performance record. He made and lost several fortunes and went bankrupt more than once.

A widely used practitioner framework for stage and trend investing. The book teaches from chart examples, not a systematic test; a 30-week MA is roughly a 150-day average.

In the original test (allocation 1973-2005) the timing portfolio earned equity-like returns with bond-like volatility and drawdowns, and cut risk in every asset class. It lags in fast rebounds and whipsaws in sideways markets; one of SSRN's most downloaded papers.

Faith and Covel report the Turtles averaged about 80% a year compounded over roughly four years; the figure is secondhand, unaudited and varied widely by trader. The rules are mechanical and testable on futures.

Buy signals were followed by higher, less volatile returns than sell signals (about 12% vs about -7% a year, before costs), which the null models could not explain. Sullivan, Timmermann and White (1999) found the best rules did not hold up after 1986.

Bollinger documented bands as an adaptive volatility framework and warned against treating a band touch as an automatic reversal signal; performance of squeeze breakouts requires market-specific testing.

Wyckoff's tape-reading framework is a foundational practitioner method, but its pattern judgments are partly discretionary and need explicit operational definitions for robust backtesting.

Windsor under Neff, 1964-1995: 13.7%/yr vs 10.6% for the S&P 500 with dividends; beat the index in 22 of 31 years. The ratios are mechanical; the cyclical and industry judgment is not.

Simon and Wiggins (2001) found put/call, VIX and TRIN had contrarian power for S&P 500 futures over 10 to 30 days in 1989-1999. For single stocks, Pan and Poteshman (2006) found low put/call names beat high put/call names by over 1% the next week, using non-public signed volume.

Alldredge and Blank (2019) found clustered purchases earned about 2.1% abnormal return the next month vs about 1.2% for solitary purchases. Kang, Kim and Wang reported 3.8% vs 2.0% over 21 trading days (1986-2016). Roughly double, not several times.

US optionable stocks, about 1996-2010: the lowest O/S decile beat the highest by 0.34% per week (about 19.3% annualized), more where shorting was costly. Weekly turnover is high and the edge is short-lived.

BBW, 1968 to 2008: $1 in the lowest-volatility quintile grew to $59.55 ($10.12 real) while $1 in the highest fell to $0.58. They blame benchmarked managers plus demand for lottery-like stocks. Low-vol lags in strong bull markets.

US equity BAB, 1926 to 2012: Sharpe ratio 0.78. The authors report positive, significant BAB returns in international stocks, Treasuries, credit and futures. Novy-Marx and Velikov (2022) find much of the US result comes from tiny stocks and the nonstandard beta estimate.

Run it: the terminal computes the verdict live against BTC.

Run it: the terminal computes the verdict live against BTC.

Run it: the terminal computes the verdict live against BTC.

Run it: the terminal computes the verdict live against the crowd. Polymarket only, about 200 bets: a small sample.

Run it: the terminal computes the verdict live against the crowd. Polymarket only, under 200 bets: a small sample.

Run it: the terminal computes the verdict live against the pile.

Run it: the terminal computes the verdict live against the pile.

Run it: the terminal computes the verdict live against every name.

Run it: the terminal computes the verdict live against every name.

Over July 1963 to July 1995 industry momentum survived controls for size, book-to-market, stock momentum and microstructure, and was strongest in the first month after formation, where stock momentum instead reverses. How much stock momentum it explains is disputed in later work.

In US stocks from 1927, the worst momentum losses came in rebounds: in July-August 1932 losers returned 232% vs 32% for winners, and in March-May 2009 losers rose 163% vs 8%. The dynamic strategy approximately doubled the alpha and Sharpe ratio of static momentum in sample, and the pattern appears in other markets.

Residual momentum earned risk-adjusted profits about twice those of total-return momentum, was more consistent over time, and was less concentrated in the extreme stocks of the cross-section in the long US sample.

The book's 1974 to 2013 backtest shows 17.43% per year with a 22.72% maximum drawdown, far shallower than the equity indexes' 2008 losses. It is an in-sample result; post-publication performance has been weaker, with whipsaw losses.

Over the 36 months after formation, losers beat the market by 19.6% and winners lagged by 5.0%, a 24.6% spread (t = 2.20). The effect was mostly on the loser side and much of it came in Januaries; later work ties part of it to size, low-priced stocks and changing risk.

In NYSE and AMEX stocks, 1962 to 1986, prior-week winners and losers reversed sharply the next week, and Lehmann's profits survived his bid-ask and cost corrections. Later studies find the effect concentrated in small, illiquid stocks; implementability depends on trading costs.

Novy-Marx found momentum is driven mainly by returns 12 to 7 months back, and that recent-return (6 to 2 months) strategies earn less, especially in large, liquid stocks, with similar results in index, commodity and currency markets. Later international tests found the pattern weaker outside the US.

Grinblatt and Moskowitz found consistent winners outperform other winners and tax-loss selling shapes the December-January pattern; their parsimonious hedged rule earned large returns in sample after controls for other premia and data snooping.

Heston and Sadka found stocks with high returns in a given calendar month tend to outperform in that same month in later years, with the pattern lasting up to 20 annual lags on top of ordinary momentum and reversal effects.

For stocks with similar formation returns, momentum fell steadily from 5.94% for the most continuous-information stocks to -2.07% for the most discrete, and continuous-information momentum showed no long-run reversal.

Over 1927 to 2011, scaling momentum by its recent realized volatility raised the Sharpe ratio from 0.53 to 0.97 and largely removed its crashes. The scaled strategy uses time-varying leverage and is shown before trading and financing costs.

The average factor earned 1 basis point a month after a losing year and 53 basis points after a positive year. The authors conclude factor momentum explains individual stock momentum, which crashes when factor autocorrelations break down.

Returns not explained by cash-flow news reversed more strongly than news-driven returns, and the enhanced reversal strategy earned a risk-adjusted return about four times that of standard short-term reversal in sample. The authors tie the long side to liquidity shocks and the short side to sentiment and short-sale limits.

Oppenheimer (1986): baskets bought each year-end 1970 to 1982 and held one year returned a mean 29.4%/yr vs 11.5% for the NYSE-AMEX index. Qualifiers are mostly tiny, illiquid stocks; few US names qualify today.

Campbell and Shiller (1998) found P/E10 forecast 10-year real price changes in US data back to 1871, but not earnings growth; power at 1-year horizons is weak. CAPE stayed above its long-run mean for most of 1990-2020, so timing rules were often under-invested.

Global high-minus-low B/M returns averaged 7.68%/yr in 1975-1995, and value beat growth in 12 of 13 major markets; a value premium also appeared in emerging markets. Premiums vary widely by country and period.

Firms with R&D earned about the same average returns as firms without; firms with high R&D relative to market value earned large excess returns; R&D intensity went with higher return volatility.

Peters and Taylor did not test returns: adding intangibles to q gives a stronger investment-q relation. Eisfeldt, Kim and Papanikolaou (2022) found an intangible-adjusted value factor earned substantially higher returns than standard HML, including in recent decades.

Classified 95% of its 66-firm sample correctly one year before failure, 72% two years before. Later work (Dichev 1998) found high-risk firms earned lower returns.

On 74 manipulators (1982-1992) it caught about half before public discovery, while also flagging some firms that had not manipulated their earnings. Cornell students reportedly flagged Enron with it in 1998.

1964-2002: lowest minus highest NOA decile earned about 1.24%/month abnormal (equal weighted) in year one, less in years two and three; profitable in 35 of 38 years.

Using a Jones-model split, Xie finds the market overestimates how long abnormal accruals persist and overprices them; they account for much of the accrual anomaly.

The authors find less reliable accrual categories have lower earnings persistence and are not fully anticipated by investors, producing documented mispricing in their sample.

The paper finds cash-based operating profitability predicts cross-sectional returns better than profitability measures that include accruals and subsumes accruals in its tests.

Mohanram reports that high G-Score growth firms outperformed low-score firms in the study, with much of the long-short result coming from the short side.

Soliman finds that DuPont components predict future earnings and that incomplete investor and analyst processing is associated with future abnormal returns.

Penman and Zhang find their diagnostics forecast future return on net operating assets and stock returns, consistent with investors not fully recognizing the interaction of conservatism and investment changes.

Fit on 105 bankruptcies and about 2,058 industrial firms (1970-1976), the logit predicts bankruptcy in sample. Re-estimate its coefficients before using it on a modern universe.

Announcers minus non-announcers earned over 60 bps a month, 7% to 18% a year across specifications, over 1927 to 2004, including in large caps. Later work reports the US premium shrank sharply after the mid-2000s.

Announcement gains were about 3% in 1976 to 1983 (2.79% in Shleifer). The effect grew to about 7.4% in the 1990s, then fell below 1% in the 2010s (Greenwood and Sammon), a clear case of decay.

Cusatis, Miles and Woolridge (146 spinoffs, 1965 to 1988) found positive abnormal returns for up to three years, concentrated in firms that later became takeover targets. Greenblatt cites about 10% a year of outperformance; his reported Gotham record of about 50% a year came from special situations broadly.

About -1.5% abnormal over three days around expiry and a permanent 40% rise in volume (1,948 lockups); larger for VC-backed firms. A small edge; borrow cost and availability limit the short.

Womack (about 1989 to 1991) found drift of +2.4% after new buys, short-lived, and -9.1% after new sells over six months. Gleason and Lee found revision drift larger for high-innovation revisions and thinly covered firms.

About 4% a year excess return after transaction costs and practical limits (4,750 deals, 1963 to 1998). Returns are uncorrelated with the market in flat or rising markets but fall with it in severe declines, like selling index puts.

+3.4% over the three announcement days and +7.5% market-adjusted over the next 12 months. The authors found initiation drift less robust than the drift after omissions.

-7.0% over the three announcement days and -11.0% market-adjusted over the next 12 months. Omission drift was stronger and more robust than initiation drift.

Announcement return of 3.38%, then 7.93% excess in the first year and 12.15% over three years (1,275 splits). Byun and Rozeff (2003) found long-run post-split returns small and sensitive to method.

1,526 IPOs from 1975 to 1984 returned 34.5% over three years from the first-day close vs 61.9% for industry and size matched firms. Later studies find the gap depends on period and benchmark and is concentrated in small issuers.

SEO firms returned about 7% a year over the five years after issue (1970 to 1990), well below size-matched nonissuers; across issuers, 44% more money was needed to end with the same wealth. Later work finds the gap shrinks with book-to-market matching and vanished for 2003 to 2012 offerings.

About 7% abnormal return around the announcement (2001 to 2006), with no reversal over the next year; funds got at least partial success in about two-thirds of cases. Part of the move comes before the filing is public.

Harris and Gurel's headline is on additions; they report deletions fell too, apparently temporarily (deletion size not re-verified here). Chen, Noronha and Singal (2004) find no permanent decline for deletions, and Greenwood and Sammon report the deletion effect near 0.1% in 2010 to 2020.

Irvine found an initiation's price impact 1.02% larger than that of a recommendation by an existing analyst; liquidity improved after initiations but did not fully explain the extra return. Post-event drift was not the paper's finding.

DeAngelo, DeAngelo and Skinner found specials were once common and paid almost as predictably as regular dividends, then largely disappeared; only very large specials survived, and repurchases did not displace them. It is a payout-policy study, not a tested trading rule.

DJIA 1897-1986: the 4 days averaged 0.473% and held all of the index's gain. CRSP 1926-2005: 0.15% a day on those days vs about 0% on the other 16; seen in 31 of 35 countries. Post-2005 strength not verified here.

Sept 1994 to Mar 2011: about 49 bp average gain per window, about 80% of annual US excess stock returns. Kurov, Wolfe and Gilbert (2021) find it essentially disappeared after 2015.

Cooper, Cliff and Gulen (SSRN 2008): the 1993-2006 US equity premium was all overnight, in stocks, indices and futures. Lou, Polk and Skouras (2019): anomaly profits split cleanly into overnight or intraday legs. Net-of-cost profits not shown.

Ariel (1963-1982): pre-holiday days averaged 9 to 14 times other days and earned over a third of the market's return. Ko and Yang (2024): in 1983-2019 it is insignificant for large firms and survives mainly in small firms.

DJIA 1897-2004: more than 90% of capital gains came on out-of-session days, about 13.8% annualized out of session vs about 2% in session. Unpublished working paper; no out-of-sample test cited.

Almanac-based tallies for the Dow since 1896: about 10% average in pre-election years vs about 3% to 6% in the other years (figures vary by edition). About 30 cycles; the no-loss pre-election streak ended in 2015.

French: S&P composite Monday returns (Friday close to Monday close) were significantly negative in each 5-year subperiod of 1953-1977, other weekdays positive. After the late 1980s the effect largely disappeared or reversed in large-cap US indices (Schwert 2003).

Ariel (1963-1981): mean returns were positive only just before and during the first half of the month and near zero in the second half. Later US data (McConnell and Xu 2008) put the gains mainly in days -1 to +3, not the rest of the first half.

In holdings of 769 pension funds ($129 billion, late 1980s), managers sold losers more heavily, especially in the fourth quarter. The paper measures trading, not a profitable public trading rule.

In nine markets, returns were lower in autumn and higher in winter as daylight changed, stronger at higher latitudes and reversed in the Southern Hemisphere. Kelly and Meschke (2010) dispute the SAD link; not an established premium.

Kamstra, Kramer and Levi: losses on clock-change weekends were about 2 to 5 times the normal weekend effect (US, Canada, UK, Germany). Pinegar (2002): significant only for the fall change and driven by two outliers.

Stivers and Sun reported relatively high expiration-week returns for option-active S&P 100 stocks and modest subsequent-week weakness; whether the pattern survives newer option-market structure is uncertain.

Lakonishok and Smidt's 1897-1986 Dow study is the cited source, but whether it singles out September is not verified. Treat September weakness as descriptive and prone to data mining.

Across 48 countries, returns were lower around full moons than new moons by about 3% to 5% a year in pooled global portfolios. Not an established tradable premium; vulnerable to multiple testing.

O'Neil states that no major bull market began without a follow-through day, but many follow-throughs fail. This is his hindsight count, not an independent test; the false-positive rate needs measuring.

The authors report high win rates and small average gains per trade on US index and stock data. With no stop and short holds, costs, gaps and slippage decide the net result.

Crabel documented the opening-range breakout as a short-term futures trading framework; its returns are highly sensitive to market, opening-window definition, execution, and intraday costs.

Elder presented Triple Screen as a discretionary but rule-specifiable multi-time-frame method, rather than as a peer-reviewed claim of universal outperformance.

Shannon presented anchored VWAP as a structured support, resistance, and trend reference; selecting anchors after the fact can create substantial discretion and backtest bias.

Kaufman documented the adaptive-average construction to address changing noise levels; any trading advantage versus fixed averages depends on parameter choices and trading costs.

Appel documented MACD as a momentum and trend indicator; crossovers can lag sharp reversals, so performance varies materially by instrument, horizon, and cost assumptions.

Wilder introduced the Directional Movement System and ADX as a systematic trend-strength framework; thresholds and smoothing periods are parameters to be tested rather than guaranteed sources of return.

Wilder presented Parabolic as a stop-and-reverse trend system intended to protect open profits; it is prone to whipsaw during range-bound markets.

1957-1969: 29.5%/yr for the partnership before Buffett's share, 23.8%/yr to limited partners, vs 7.4% for the Dow with dividends; no losing year. Small capital made it possible.

No documented record exists for Fisher. Buffett said in 1969 he was '85% Graham, 15% Fisher' and later moved toward Fisher; the points underpin quality-growth research.

The book reports hypothetical indicator results, not an audited record. Zweig warned of a crash days before October 19, 1987; the model's live record is not public.

Baupost (founded 1982) is private and its record is not public; press reports cite high-teens yearly returns, unverified. The book is out of print and resells at collector prices.

A framework, not a rule: Marks gives no mechanical signal or backtest, and Oaktree's fund records are only partly public. Known for adding distressed-debt risk in busts such as 2008.

Quantum is widely reported at about 30%/yr from 1969 to 2000, with losing years. The 1992 sterling short, led by Druckenmiller, made about $1 billion. The theory is not a testable rule.

Duquesne reportedly averaged about 30%/yr with no losing year until he returned outside money in 2010; the records are private. He also lost heavily on tech in 2000 at Soros.

The 1939 basket of about 104 sub-$1 stocks roughly quadrupled in four years. Templeton Growth Fund is widely cited at about 14.5%/yr from 1954 to 1992.

The best-known risk-parity portfolio. Bridgewater's own backtests show equity-like returns with less volatility; it had its worst year in 2022 when stocks and bonds fell together.

APR (1988-2002): constrained stocks trailed by 2.15%/month equal-weighted but only 0.39%/month value-weighted, not significant. BJZ (NYSE, 2000-2004): heavily shorted stocks trailed lightly shorted ones by 1.16% over the next 20 trading days. The effect sits in small, hard-to-borrow names.

Wang, Yan and Zheng (2020) found disclosed shorting flows still predicted lower returns in 2010-2015, with lightly shorted stocks beating heavily shorted ones. Single-name readings are noisy. In our warehouse from 2019.

US stocks, about 1980-2003: high-flow stocks trailed low-flow stocks; three-year flow spreads were about 0.4% to 0.85% per month depending on risk adjustment. Much of the effect overlaps value and share issuance.

Wermers (1975-1994): stocks herds bought beat stocks they sold by about 4% over six months, mostly in small stocks. Coval and Stafford (1980-2004): fire-sale stocks fell and recovered gradually over more than a year. Overlaps the Coval-Stafford entry.

US optionable stocks, 1996-2005: the steepest-smirk quintile trailed the flattest by 10.9% per year risk-adjusted, and the gap lasted at least six months. Steep-smirk firms had the worst next-quarter earnings news. Later periods need separate testing.

US optionable stocks, 1996-2005: stocks with relatively expensive calls beat those with relatively expensive puts by about 50 basis points per week, with both legs contributing. The predictability shrank over the sample period.

Markit fee data, 2004-2012: the cheap-minus-expensive-to-short portfolio earned about 1.43% per month gross and about 0.91% net of lending fees. Major anomalies were weak in the 80% of stocks with low fees and strong in high-fee stocks. Working paper; numbers vary by version, and the short leg needs institutional borrow access.

TAQ, 2010-2015: stocks with net retail buying beat stocks with net selling by about 10 basis points the next week, with some predictability out to 12 weeks. Barber et al. (2024) found the method catches only about 35% of retail trades and mis-signs about 28% of those.

US stocks, 1963-2001: after high sentiment, small, young, volatile, unprofitable, non-paying and distressed stocks earned relatively low returns; after low sentiment the same spreads turned positive. The index is estimated on the full sample, so it is not real-time.

NYSE and sampled Nasdaq stocks, about 1993-2002: stocks with no coverage in four national newspapers beat highly covered stocks by about 3% per year after risk adjustment, and about 8% to 12% a year among small, low-analyst, retail-held stocks. Newer media sources may change the effect.

Russell 3000, 2004-2008: a rise in abnormal search volume predicted higher prices over the next two weeks and a reversal within the year. For IPOs, higher search attention went with larger first-day returns and weaker long-run returns. Short sample; Google data is sampled, so pulls differ.

US closed-end stock funds, about 1965-1985: discounts moved together and narrowed when small stocks did well, consistent with a shared sentiment factor. The paper did not test a trading rule, and later work disputed the small-stock link. Turning this into a convergence trade is fund-specific.

US 1928-1997: a high equity share preceded low next-year market returns and beat dividend yield and book-to-market in sample. Butler, Grullon and Weston (2005) argued much of the fit reflects pseudo market timing and a few early years. Annual, slow signal.

Miller supplied the theory: with short-sale limits, disagreement lifts prices. It is not an empirical trading test. Later, Diether, Malloy and Scherbina (2002) found stocks with high analyst forecast dispersion earned lower returns, mostly in small stocks.

US mutual funds, 1980-2004: stocks under heavy selling by extreme-outflow funds (bottom 10% of flow pressure) fell and then recovered gradually over more than a year; inflow-driven buying showed the mirror pattern. Holdings arrive with a lag, so timing is hard.

Bakshi and Kapadia find negative delta-hedged gains on S&P 500 options, a negative volatility risk premium. Put-writing indexes show the premium but take large losses in crashes such as 2008 and March 2020. The Feb 2018 blowup hit short-VIX notes, a different trade.

Simon and Campasano report that the basis predicts VIX futures returns, not spot VIX changes. On Feb 5 2018 the short-VIX note XIV lost about 96% of its indicative value in one session and was terminated.

The spread inverted before each US recession from 1969 to 2020, but the 2022 to 2024 inversion (low of -1.89 pt) had no recession through Aug 2026. Our test, S&P 500 TR only when the spread is positive, 1989 to 2026: 9.9% a year vs 11.4% buy and hold.

Asness finds the Fed model explains how investors priced stocks against bonds since the 1960s but does not forecast 10-year real stock returns; starting P/E alone does better. Kept as the canonical plausible-but-wrong signal.

Our data: log gold vs the 10-year TIPS yield correlated -0.92 monthly over 2006 to 2021 but +0.53 over 2022 to Sep 2026, as gold rose from about $1,640 to $5,250 with real yields near 2%. Monthly moves still ran inverse (-0.37). Central-bank buying is a common, unproven explanation.

FRED real-time data: fired 2 to 4 months after the cycle peak in 8 of 9 recessions since 1960 (1969: 2 months before). Readings of exactly 0.50 in 1976 and 2003 had no recession. Fired July 2024 (0.53) with no recession through Aug 2026.

HML FX and a dollar factor explain most of the spread across carry portfolios in 1983 to the late 2000s. Carry is exposed to global volatility shocks and lost heavily in 2008.

Winner-minus-loser spreads of up to about 10% a year (1976 to 2010, before costs), weakly related to carry. Profits sit mainly in minor, high-risk currencies where trading costs and limits to arbitrage eat much of them.

Gorton and Rouwenhorst (1959 to 2004) find an equal-weight, collateralized futures index earned an equity-like premium. The basis sort comes from later work: Gorton, Hayashi and Rouwenhorst link backwardation to low inventories and find backwardated contracts earned more. Not a guarantee for any single commodity or period.

Erb and Harvey find single futures averaged near-zero excess returns; roll return explained most differences across commodities, and rebalancing added a diversification return. An equity-like strategic premium is unproven.

Premia found in all eight markets (1972 to 2011), with value and momentum negatively correlated within and across asset classes and linked to funding liquidity risk. Realized premia vary a lot over time.

Across eight US inflation regimes since 1926, stocks and nominal Treasuries lost in real terms on average while commodities and trend-following gained. With only eight regimes, the estimates are noisy.

Faber reports favorable historical results for allocating to lower-CAPE country markets, but country valuation can remain extreme and the signal can underperform for extended intervals.

In 1964 to 2003 data one factor explains up to about 44% of the variation in one-year excess returns on 2 to 5 year bonds, far more than yield-spread models. Out-of-sample results in later studies are weaker and sensitive to estimation choices.

The framework became influential among FX practitioners as a regime description, but its simple rules should be treated as partial and validated out of sample rather than assumed stable.

Browne presented the allocation as a robustness-oriented investment plan rather than a peer-reviewed return anomaly; results depend on the historical behavior of its four sleeves and may differ materially going forward.