Trading strategies built and stress-tested on our own backtesting system, each with its full report. Every one published here failed. The ones that work, we keep.
Every strategy on this page failed. That is the point.
We publish these to show how our system tests an idea and how its reports read, not to sell the ideas. Each one looked plausible, was built properly, and did not hold up. The strategies that do work are not on this page, and never will be.
What you are seeing
Real strategies, run through the same backtesting system and report generator we use on our own ideas. Every chart, every trade and every caveat, unedited.
What you are not seeing
Anything that works. A strategy that survives our tests stays private. The ones published here are the ones that did not make it.
What this is not
Investment advice, a signal service or a track record. These are historical simulations of rules that failed. Do not trade them.
The lab notebook
The Strategies
One row per strategy: the rule, what it did trade by trade, and why it does not work. Open any of them for the full report.
All trades: 818 trades, 415 winners (51%). Each line runs from the day a stock was bought to the day it was sold; returns are on a log scale, so −50% and +100% sit the same distance from zero.
At every month-end, rank the S&P 500 as it stood at the end of 2021 on each stock's return over the past year, skipping the latest month, and hold the 50 best in equal weight until the next ranking.
12-month return, skip 1
Top 50 of 455
Monthly rebalance
Equal weight
No stop
S&P 500 as of 2021
5 bp a side
Why it fails
It beats SPY from four of five start dates, and loses to it in most of the windows you could actually have picked.
From 2022: 14.3% a year against SPY's 12.2%, but at a lower Sharpe (0.72 against 0.75) and a slightly deeper worst fall (−24.7% against −24.5%). More return for more risk, not a better deal.
Beat SPY in only 42% of 914 one-year windows and 32% of 410 three-year windows. Against QQQ over three years: 2%.
The lead rests on 2022's smaller loss and the first eight months of 2026 (+32.3% against +12.9%). In 2023 it made +6.6% against SPY's +26.2%.
50 of the 505 names have no prices, mostly ones acquired or failed since 2021, so it trades the survivors: a tilt in its favour the report flags. Nothing here is out of sample.
All trades: 843 trades, 434 winners (51%). Each line runs from the day a stock was bought to the day it was sold; returns are on a log scale, so −50% and +100% sit the same distance from zero.
Momentum scored only on the part of each stock's past-year move that the rest of the S&P 500 doesn't explain, so a high-beta tech name and a consumer staple compete on equal terms. Hold the top 50, re-ranked monthly.
Residual 12-1 return
Beta window 252 days
Top 50 of 455
Monthly rebalance
No stop
S&P 500 as of 2021
5 bp a side
Why it fails
Level with SPY on risk-adjusted return from 2022, and behind it in most windows.
From 2022: 12.5% a year against SPY's 12.2%, at the same Sharpe (0.75). A shallower worst fall (−23.0% against −24.5%), but no more return per unit of risk.
Start a year later, in 2023, and it trails SPY by 2.7 points a year (19.9% against 22.6%).
Beat SPY in 42% of one-year windows and 36% of three-year windows; QQQ in 2% of the three-year ones.
2023, the rebound a beta-neutral score should have handled differently, was +8.5% against SPY's +26.2%, barely better than plain momentum's +6.6%.
All trades: 846 trades, 426 winners (50%). Each line runs from the day a stock was bought to the day it was sold; returns are on a log scale, so −50% and +100% sit the same distance from zero.
Each month, keep the half of the S&P 500 whose past-year move came as a steady drip of small days rather than a few big jumps, and hold the 50 with the best 12-1 return among them, on the theory that slowly arriving news is the news markets underreact to.
Information discreteness
Keep smoothest 50%
Rank on 12-1 return
Top 50 of 455
Monthly rebalance
No stop
5 bp a side
Why it fails
Behind SPY from 2022, and in roughly two of every three one-year windows.
From 2022: 11.9% a year against SPY's 12.2%, at a Sharpe of 0.66 against 0.75.
Beat SPY in 35% of 914 one-year windows and 27% of three-year windows; QQQ in 3% of the three-year ones.
2023 was +6.7% against SPY's +26.2%. Filtering for smooth winners did nothing for plain momentum's bad year (+6.6%).
Its five biggest contributors are plain momentum's five, WDC, STX, HWM, MU and GE, so the filter mostly reshuffles the same bet.
All trades: 693 trades, 335 winners (48%). Each line runs from the day a stock was bought to the day it was sold; returns are on a log scale, so −50% and +100% sit the same distance from zero.
Rank the S&P 500 on what each stock gained overnight only, from each close to the next open, over the past year, and hold the top 50 for a month. The trading-hours half of every day's move is thrown away.
Overnight 12-1 return
Top 50 of 455
Monthly rebalance
Equal weight
No stop
S&P 500 as of 2021
5 bp a side
Why it fails
The deepest fall and the lowest Sharpe of the set, and it beat SPY in only a quarter of one-year windows.
From 2022: 11.1% a year against SPY's 12.2%, at a Sharpe of 0.59 against 0.75, with a worst fall of −27.8% against −24.5%.
Beat SPY in 26% of 914 one-year windows and 21% of three-year windows.
Its worst three-year window lost 4.9% a year. SPY's worst still made 4.6%.
One trade, STX held since August 2024 and still open, is up 759%. A result that leans on one position is a result about that stock.
All trades: 1092 trades, 576 winners (53%). Each line runs from the day a stock was bought to the day it was sold; returns are on a log scale, so −50% and +100% sit the same distance from zero.
Score every S&P 500 stock on a hand-set blend of 60% past-year momentum, 30% low volatility and 10% last-month reversal, and hold the 50 highest scores, re-ranked monthly. It is the seed our genetic search starts from.
60% momentum
30% low volatility
10% reversal
Top 50 of 455
Monthly rebalance
No stop
5 bp a side
Why it fails
The shallowest fall of the five ranking rules, paid for with the worst record: it beat SPY in 1% of three-year windows.
From 2022: 11.6% a year against SPY's 12.2%. The low-volatility tilt held its worst fall to −21.2% against SPY's −24.5%, and its 2022 loss to 7.2% against 18.6%.
Then it gave the lead back: +5.1% in 2023 against SPY's +26.2%, and +5.9% in 2025 against +17.7%.
Beat SPY in 18% of one-year windows and 1% of 410 three-year windows; QQQ in none of the three-year ones.
The most turnover of the five, 481% a year one way, so it pays the most in costs.
