We tested the famous strategies in our library. Here's how many still work.

// The short answer

Most of the famous stock strategies beat the market from 1998, and most of them have trailed it since 2015. Of the 42 US stock strategies we ran over the full window, 26 grew faster than the S&P 500 with dividends from January 1998 (the index grew 9.38% a year to September 2026, gross of costs). From January 2015, when the index grew 14.17% a year, 4 did.

Our engine gives each run one of three findings. A strategy that beat its benchmark on growth and on Sharpe ratio (return for the risk taken) and did not fall behind after a split at 2015 reads DATA VALIDATED. One that won on only one of the two, or won on both over the whole run and then trailed on both after 2015, reads TRADE-OFF. One that trailed on both reads NO EDGE. A finding describes what the data did over the window; it is not a prediction.

• US stock strategies44of 55 runnable entries
• Data validated3and 2 outside US stocks
• Trade-off29and 1 outside US stocks
• No edge12and 4 outside US stocks

The 3 US stock strategies that came out DATA VALIDATED: Quantitative Value (inspired by Gray and Carlisle) at 17.5% a year against 9.4%; Quality Minus Junk (inspired by Asness, Frazzini and Pedersen) at 16.8% a year against 9.4%; Cross-Sectional Momentum (inspired by Jegadeesh and Titman) at 16.5% a year against 9.4%. Of the other runnable entries, 7 run outside US stocks against their own benchmarks, 2 are insider-filing event studies that give no finding by design, and 2 did not run: Memecoin 30d Momentum, Graveyard Included (the memecoin graveyard holds 2 months of history; a run needs 13); Memecoin Survivors, 180 Days and Volume (the memecoin graveyard holds 2 months of history; a run needs 13). These are backtests: history, not a forecast, and 55 tests side by side will turn up some winners by chance alone.

// The count

Of the 44 US stock strategies, 3 came out DATA VALIDATED, 29 TRADE-OFF and 12 NO EDGE. The 7 entries outside US stocks (crypto, prediction markets and the 24/7 tape) are each scored against their own benchmark, so they get their own bar here and their own table below. By library category:

// Every strategy against the index

Over the whole run from 1998, 26 of the 42 grew faster than the index, 23 had a higher Sharpe ratio and 14 fell less at their worst month-end. The fastest was Quantitative Value (inspired by Gray and Carlisle) at 17.45% a year, with a worst fall of -51.1%. The slowest was Short-Horizon Reversal (inspired by Jegadeesh and Lehmann) at -1.30% a year, a loss.

The chart sets each strategy's lead or lag against the index before 2015 beside its lead or lag from 2015. 29 of the 42 dots sit to the right of zero and below it: ahead from 1998 to 2014, behind from 2015 to 2026. 3 are ahead in both parts.

// Before and since 2015

The index itself changed pace: 6.21% a year before 2015, 14.17% from 2015. 14 of the 29 TRADE-OFF results are the same shape, ahead on growth and Sharpe over the whole run and behind on both after 2015. Value was the clearest case: 12 of the 13 value books on the standard window beat the index before 2015 and 1 from 2015, when the median value book grew 12.65% a year. Our books hold 20 or 30 names at equal weight; the index weights its largest companies most. 4 lost money from 2015: Momentum as a Fourth Factor (inspired by Carhart) at -4.87% a year; Short-Horizon Reversal (inspired by Jegadeesh and Lehmann) at -0.60% a year; Band Squeeze and Breakout (inspired by Bollinger) at -0.15% a year; Stage Analysis (inspired by Weinstein) at -0.13% a year.

The split is not a true holdout. 41 of the 42 sources were published before 2015, and their ideas were public through the later years, so those years test a known idea, not a blind one. Left out of this section: RSI(2) Mean Reversion (inspired by Connors and Alvarez) (Jan 2016 to Oct 2026) and Option-to-Stock Volume (inspired by Johnson and So) (Jun 2014 to Sep 2026), whose data starts later.

// Every run, US stocks

Grouped by finding, then sorted by how far each one finished ahead of or behind the index on growth. Red is a figure below the index's over the same months, or a loss. Each name opens its library entry, with the source's own claim beside our criteria.

// Outside US stocks

7 entries ran outside US stocks, each on its own engine, against its own benchmark and over its own window, so these rows do not compare with the S&P 500 rows above. All are gross of fees.

• Crypto, against Bitcoin bought and held. Crypto 12-1 Month Momentum (Binance) 10.4% a year against 35.9%; Crypto Low Volatility 19.0% a year against 27.4%; Crypto Trend + Strength -15.5% a year against 27.6%. All 3 trailed Bitcoin on growth from 2018; their worst falls ran from -77.2% to -98.0%.
• Prediction markets, against every resolved market at the same timing and sizing. Prediction Market Longshots (Polymarket only) 71.6% a year against 57.4%, DATA VALIDATED; Prediction Market Favorites (Polymarket only) 6.0% a year against 57.4%, TRADE-OFF. Polymarket only: 33 monthly periods from Mar 2023, with 763 and 627 bets, a small sample for a finding.
• The 24/7 tape, against every name held 09:00 to the close. Follow the 24/7 Tape: Buy What It Bid Up Overnight -9.9% a year against -0.9%, NO EDGE; Fade the 24/7 Tape: Buy What It Sold Off Overnight 11.6% a year against -0.9%, DATA VALIDATED. 166 sessions from March 3, 2026, all after the engine's split, so each finding rests on one short window.

// Documented, not run

120 of the 175 entries are documented in the library but do not run in one call. Some need data we do not hold yet, some are long-short or event designs a one-call backtest does not take, and some cannot be traded as published; the library states the reason on 9 of them, shown beside the name. They are listed here so the count has nothing missing from it. None of them is in any number on this page.

// Method and limits

• The runs. One run_library_strategy call per runnable entry over QuantGPT's MCP on October 28, 2026, with the entry's own library defaults: its filters, ranking, book size and cadence. Every call set charts: false and wait: true, which change no number, except Insider Buying Signal and Insider Cluster Buys, whose engine refused both, so they ran with none. The 5 index timing rules (10-Month MA Timing (GTAA), Golden Cross / Death Cross, Moving-Average Rules (inspired by Brock, Lakonishok and LeBaron), Sell in May (Halloween Indicator) and Santa Claus Rally) default to 1960 on the S&P 500 price index; we set start_year: 1998 and benchmark: $SPXTR, which puts the rule itself on the total return index too. Each run's exact arguments are in the CSV.
• Window. The US stock books form at each month-end close from January 30, 1998 to August 31, 2026, so their returns run to September 30, 2026; the timing rules run to October 2026. RSI(2) Mean Reversion (inspired by Connors and Alvarez) (daily, Jan 2016 to Oct 2026) and Option-to-Stock Volume (inspired by Johnson and So) (monthly, Jun 2014 to Sep 2026) run on their own windows, named in the table. Prices were current through October 27, 2026.
• Benchmark. The S&P 500 total return index ($SPXTR), dividends reinvested, bought and held over the same months. The entries outside US stocks use their own engine's benchmark.
• Findings. The engine's own: growth a year and Sharpe ratio against the benchmark over the whole run, then again on each side of a walk-forward split, January 1, 2015 for the US stock runs. Sharpe is the mean period return over its standard deviation, annualized, with no T-bill rate subtracted, so a rule that sits in T-bills most of the year (the Santa Claus rally) shows a high Sharpe and low growth.
• Costs. Every figure is gross of trading costs and taxes except the 10 bps column, which charges 10 basis points a side on each book's measured turnover (the engine's cost model). 23 of the 39 US stock books with a cost model still grew faster than the index at that cost. The timing rules and the engines outside US stocks report no cost model.
• Delisted companies. The market-wide universe holds companies that later delisted from 1998: 12,815 of the 13,192 delisted operating companies with prices from 1998 (97%) can enter it, and the rest cannot, so results still lean a little toward survivors. Before 1998 it holds survivors only, which is one reason these runs start in 1998. A delisted holding exits at its last trade, or at -30% if it never traded again.
• One engine. Each run is the closest our factor vocabulary gets to its source, not the source's own test: mostly a long-only, equal-weight book of 20 or 30 liquid US stocks re-formed monthly, where many papers sorted deciles, went long and short, or held for a year. A name that starts "Inspired by" follows its source loosely. Each library entry lists what the run changes.
• Overlap. The strategies are not independent. Many value screens rank on close cousins of the same ratio (earnings, cash flow or book against price), and 4 pairs of entries came back with identical numbers in this engine: The Acquirer's Multiple (inspired by Carlisle) and Low Price-Earnings Portfolios; Earnings-Surprise Momentum and Post-Earnings Drift (SUE) (inspired by Bernard and Thomas); Cup-with-Handle Breakout (inspired by O'Neil) and Pivotal Points (inspired by Livermore); Golden Cross / Death Cross and Moving-Average Rules (inspired by Brock, Lakonishok and LeBaron).
• Many tests. 55 runs side by side will turn up some winners by luck. No run here was tuned to its result or re-run to improve it.
• Disclosures. The conventions behind every QuantGPT backtest are on the disclosures page.

// Run your own

• Ask your AI. Connected to QuantGPT over MCP, ask:

  Use QuantGPT to run the Gray and Carlisle Quantitative Value entry from the strategy library from 1998 against the S&P 500 with dividends, then show me every calendar year it trailed and what it did after 2015.

  Swap in any entry, change its rules, and run it again.
  Connect Claude · Connect ChatGPT
• The strategy library. Every entry on this page, with the source's own claim, its criteria and what our run changes.
  Browse the library
• One strategy in depth. Every year, the split at 2015 and the variants, for one rule at a time.
  The 200-day moving average · The Piotroski F-Score
• The data. Every run on this page, one row each, with a source line.
  Download the CSV
• New to QuantGPT? 3 days free on QuantGPT or QuantGPT Pro, cancel any time.
  Start your 3-day free trial

// Questions

Do the famous stock strategies still work?

Most beat the market from 1998 and most have trailed it since 2015. We ran every runnable strategy in QuantGPT's library on October 28, 2026. Of 42 US stock strategies with the full window, 26 grew faster than the S&P 500 with dividends from January 1998 to September 2026 (the index: 9.38% a year, gross of costs), and 4 did from January 2015 (the index: 14.17% a year). Of all 44 US stock strategies, 3 were DATA VALIDATED, 29 TRADE-OFF and 12 NO EDGE.

Which strategies held up after 2015?

4 of 42 grew faster than the S&P 500 with dividends from January 2015 in our runs, gross of costs: Quality Minus Junk (inspired by Asness, Frazzini and Pedersen), 17.50% a year against 14.17%; Quantitative Value (inspired by Gray and Carlisle), 16.22% a year against 14.17%; Cross-Sectional Momentum (inspired by Jegadeesh and Titman), 16.10% a year against 14.17%; CANSLIM (inspired by O'Neil), 14.18% a year against 14.17%.

Which strategies had no edge?

12 US stock strategies trailed the S&P 500 with dividends on growth and on Sharpe: Short-Horizon Reversal (inspired by Jegadeesh and Lehmann), -1.30% a year against 9.38%; Band Squeeze and Breakout (inspired by Bollinger), -0.08% a year against 9.38%; 52-Week High Momentum (inspired by George and Hwang), 0.03% a year against 9.38%; Momentum as a Fourth Factor (inspired by Carhart), 2.22% a year against 9.38%; Stage Analysis (inspired by Weinstein), 2.31% a year against 9.38%; Turtle Breakout (inspired by Dennis and Eckhardt), 3.02% a year against 9.35%; RSI(2) Mean Reversion (inspired by Connors and Alvarez), 4.40% a year against 15.17%; Piotroski F-Score, 4.92% a year against 9.38%; Cup-with-Handle Breakout (inspired by O'Neil), 5.17% a year against 9.38%; Pivotal Points (inspired by Livermore), 5.17% a year against 9.38%; Option-to-Stock Volume (inspired by Johnson and So), 7.20% a year against 13.71%; CANSLIM (inspired by O'Neil), 7.71% a year against 9.38%. Short-Horizon Reversal (inspired by Jegadeesh and Lehmann) and Band Squeeze and Breakout (inspired by Bollinger) lost money over the whole run.

Did value investing stop working after 2015?

In our runs it stopped beating the index. 12 of the 13 value books from January 1998 beat the S&P 500 with dividends before 2015; 1 did from January 2015, when the index grew 14.17% a year and the median value book 12.65%. Over the whole run most value books were still ahead, which is why most read TRADE-OFF rather than NO EDGE.

How were the strategies tested?

One run_library_strategy call per runnable entry over QuantGPT's MCP on October 28, 2026, each with its library defaults, from the January 30, 1998 close to September 30, 2026 for most US stock books, against the S&P 500 total return index, gross of costs, with a walk-forward split at 2015. The stock universe holds delisted companies from 1998 (97% of those with prices). Every run's arguments and numbers are in the free CSV.