Open · formula shown

gen0194

A rules-based equity selection strategy. This was the first strategy discovered by the engine, and it is tracked here as an append-only paper portfolio.

Disclaimer

This is a simulated paper portfolio — no real money is being traded here. This is not investment advice.

Starting Capital: $1,000,000
Rebalance: 42 trading days
Costs: 5/5 bps
Live since: Jun 15, 2026

Plain English

Thesis, behaviour & risks

Thesis. Combine several independent price, volume, trend, and mean-reversion signals to rank a broad liquid US equity universe, then hold the strongest candidates on a fixed review schedule.

Expected behaviour. The portfolio is concentrated and can differ materially from the index. It should participate when its selected signals persist, while periodic reviews let the basket adapt as leadership changes.

Risks

Concentration and factor crowding can create sharp drawdowns.
Signal relationships learned from historical data may decay out of sample.
Daily-bar paper fills, liquidity estimates, and market-impact assumptions can differ from executable prices.

Deep analytics →Annual returns, rolling Sharpe, drawdown anatomy, factor & sector exposure, capacity, and the full rebalance history.

Lifecycle

Training → out-of-sample → live

Capacity-capped account equity

The line is total simulated account value, shaded by phase: the in-sample training window(s) it was fit on, the held-out out-of-sample window, and live paper-trading after the marker. Per-phase stats are in the panels below.

Showing 76 observations from Jun 15, 2026 to Oct 1, 2026: -11.0% total return and -11.0% maximum drawdown in this view.

Live (paper)S&P 500Log scale

Drawdown

Download live data Download full-resolution research data

Full history is a continuous account: above the estimated capacity ($36,076,014), unsupported capital remains cash. Training and out-of-sample presets use their standalone replay curves so they match the statistics shown below.

Live

Forward paper-trading

Live (paper)

since Jun 15, 2026

Real forward tracking since the live date using the Yahoo Finance API, and the same cost model as the backtests.

CAGR: -32.5%
Sharpe: -2.82
Max DD: -11.0%

Backtest

Training, out-of-sample & combined

Three deterministic single-seed runs of the frozen formula: the in-sample training window, the held-out out-of-sample window, and the two combined. Combined figures are computed end to end, not stitched from the halves.

Out-of-sample

Jan 2, 2020 – Dec 1, 2025

Never used to train the strategy, the strongest evidence short of live tracking.

CAGR: +29.6%
Total return: +363.0%
Volatility: 38.5%
Sharpe: 0.86
Calmar: 0.61
Max DD: -48.7%
Max DD length: 260d
Best year: +77.8%
Worst year: -2.4%
Worst 3y CAGR: +5.5%
Worst 5y CAGR: +26.4%
Min rolling Sharpe: 0.35
Beta vs S&P 500: 1.02
Corr vs S&P 500: 0.55
Alpha (ann.): +17.4%
Information ratio: 0.53

Training (in-sample)

Jan 1, 2000 – Jan 1, 2020

Regime 1 · 2000–2005 · Regime 2 · 2005–2010 · Regime 3 · 2010–2015 · Regime 4 · 2015–2020

The strategy was trained on this period, good returns here are not inherently indicative of quality or future performance.

CAGR: +21.5%
Total return: +4771.9%
Volatility: 25.5%
Sharpe: 0.90
Calmar: 0.37
Max DD: -57.7%
Max DD length: 798d
Best year: +74.2%
Worst year: -40.1%
Worst 3y CAGR: -2.6%
Worst 5y CAGR: +8.0%
Min rolling Sharpe: 0.08
Beta vs S&P 500: 1.20
Corr vs S&P 500: 0.90
Alpha (ann.): +14.5%
Information ratio: 1.29

Combined · training + OOS

Jan 1, 2000 – Dec 1, 2025

Combined figures are computed end to end, not stitched from the halves.

CAGR: +19.8%
Total return: +10634.9%
Volatility: 24.9%
Sharpe: 0.85
Calmar: 0.34
Max DD: -57.7%
Max DD length: 800d
Best year: +74.1%
Worst year: -40.2%
Worst 3y CAGR: -2.6%
Worst 5y CAGR: +6.7%
Min rolling Sharpe: -0.05
Beta vs S&P 500: 1.04
Corr vs S&P 500: 0.81
Alpha (ann.): +11.3%
Information ratio: 0.79

Capacity

Capacity & holdings

Active share: 40.0%
Capacity: $36,076,014

Formula

How it picks stocks

Selection: Top 28; Names are sorted by score; highest scores enter the basket.
Cadence: 2 months; The formula is re-evaluated on each rebalance date.
Inputs: 7 indicators; 3 transforms shape those raw inputs.
Exit: Has rule; A separate gate can force stale holdings out.

Each rebalance, every eligible stock is scored by the expression below. The top 28 highest-scoring names are held, weighted by rank (higher score → larger weight, each capped), refreshed about every 2 months. The exact formula is published verbatim — nothing is hidden for open strategies.

Point-in-time. Every indicator is computed through the prior trading day, so a rebalance never peeks at the bar it trades on, preventing look-ahead.
Cross-sectional vs. time-series. rankcompares a name against every other eligible name that day; z‑score standardises a value against its own recent history.
Eligible means a raw price ≥ $10, trailing median dollar volume ≥ $5M, and recent, non-stale data — the same liquidity screen the backtest used.
Comparisons and logic act as 1 / 0 gates (shown in [ ] brackets) that switch parts of the score on or off. The score's absolute value is meaningless; only the ordering across names selects the basket.

Full expression

rank₄₅₀(Dollar volume)×|((z‑score₆₀(Close-in-range)×(1.16×z‑score₅(Dollar volume)))+Realized vol₄₅₀)−(1.25÷rank₁₀(EMA₁₀₀₀))|

Exit rule

Independently of the score, this gate can force a current holding out. If the score still ranks that name back into the target basket, the score wins.

[Beta₅>(((Mean reversion₅+z‑score₅(EMA₅))×0.441)+(rank₁₀(Close-in-range)×log(Volume surge₁₂₅₀)))]

Indicators used

Beta: Rolling sensitivity of returns to the market proxy (needs a benchmark).
Close-in-range: Where the close sits within the day's high–low range (0–1).
Dollar volume: Adjusted close × share volume — a liquidity measure.
EMA: Exponential moving average of adjusted close (span = window).
Mean reversion: Price relative to its window moving average, minus one.
Realized vol: Standard deviation of daily returns over the window.
Volume surge: Volume relative to its window average, minus one.

Transforms used

log: Signed log compression: sign(x) · ln(1 + |x|).
rank: Cross-sectional percentile rank (0–1) of the feature's window-day average, across all eligible names that day.
z‑score: Time-series z-score: (value − window mean) ÷ window std — standardised against its own recent history.

Audit trail

Rebalance timeline

Append-only review, target, next-open fill, and cost events. Short hashes identify the point-in-time universe and price inputs used for each decision. Due to the fact that this was the first strategy, the first few review dates do not match the intended strategy cadence.

Sep 2, 2026correction proposed · correction accepted · basis rebased
Review recorded
Aug 26, 2026correction proposed · correction accepted
Review recorded
Aug 21, 2026correction accepted
Review recorded
Aug 11, 2026fills applied · costs charged
Review recorded · 55 fills · $486,183 estimated costs
Aug 10, 2026rebalance reviewed · targets computed
55 targets
universe 146dbc8dfc0f · prices 081b749b1db7

Composition

Current allocation

Current account allocation: 31.9% invested, 68.1% cash / uninvested.

ALM1.3%
NVDY1.3%
SAN1.2%
SVM1.2%
LPTH1.2%
NXE1.2%
AGNC1.2%
CX1.1%
ERAS1.1%
PSNL1.1%
CRVS1.1%
UEC1.0%
NOK1.0%
OSS1.0%
BB0.9%
NVTS0.9%
RCAT0.9%
TRVI0.9%
EQX0.8%
PTEN0.8%
VISN0.8%
UMC0.7%
ERIC0.7%
TTI0.6%
AMPX0.6%
WULF0.6%
AXTI0.5%
POET0.5%
BGC0.5%
LWLG0.5%
ONDS0.4%
QUBT0.4%
BW0.4%
PRCH0.4%
IAG0.3%
BKD0.3%
APLD0.2%
UMAC0.2%
CIFR0.2%
RDW0.2%
QBTS0.2%
ASX0.2%
HIMX0.2%
HBM0.1%
CORZ0.1%
HL0.1%
JOBY0.1%
RGTI0.1%
SBSW0.1%
UUUU0.1%
AG0.1%
PBI0.1%
PL0.0%
CDE0.0%
VALE0.0%
Cash / uninvested68.1%

See the full formula ↑