V3 engine · measured on 2014–2026 · costs included

We backtested every signal.
Here are the real numbers.

Most signal providers hide behind a black box. We did the opposite. Every signal comes from a 100% quantitative engine with no optimized parameter — the momentum horizons are fixed in advance, so there is nothing to overfit. Below are the actual measured results per asset class, transaction costs included, plus the strategies we tested and threw away.

1.36
Portfolio Sharpe (equal weight)
-10.4%
Portfolio max drawdown
1.22
Best single asset (Solana)
12
Years of history

Four strategies, one per asset class

A single model applied to everything is how most systems fail. Crypto trends, oil mean-reverts and shorts pay, indices simply cannot be timed. So each class gets the strategy its own data supports — and nothing else.

Trend — long or cash

Crypto (BTC, ETH, SOL, BNB)

Multi-horizon momentum (21/63/126/252d). Shorting is forbidden here: it structurally destroys value on assets with positive drift.

Trend — long or short

Oil (WTI, Brent)

The only asset class where shorting pays: WTI buy & hold is −0.21 Sharpe, the strategy turns it into +0.47.

Managed exposure 60–100%

Indices + Gold

No timing model beats buy & hold on indices — so we do not pretend to. Exposure is scaled by volatility and the 200-day average to cut drawdowns instead.

Trend on total return

US Treasuries

Trend applied to total return (coupons included), long or cash. Modest standalone, valuable as portfolio diversification.

Measured results per asset

Strategy Sharpe versus simple buy & hold, computed on 42-day rolling windows over the full available history. Nothing here is fitted: the four momentum horizons (21/63/126/252 days) are fixed before testing, so every result is out-of-sample by construction.

AssetClassBuy & holdStrategy SharpeTotal returnMax DD% positive windowsWindows
SolanaCrypto0.981.22+388%-20.8%35%48
BitcoinCrypto0.911.20+4718%-31.0%43%97
BNBCrypto0.810.87+576%-27.2%35%69
EthereumCrypto0.520.72+264%-27.4%38%69
GoldCommodities0.690.71+148%-16.2%54%71
NasdaqIndices0.800.67+216%-30.7%63%71
S&P 500Indices0.740.62+175%-33.1%65%71
NikkeiIndices0.690.59+153%-32.1%59%69
WTI OilCommodities-0.210.47+136%-28.7%42%71
US 10YBonds0.240.42+26%-8.1%39%71
Brent OilCommodities0.160.41+72%-30.8%42%71
DAXIndices0.550.40+79%-38.2%64%72
Euro Stoxx 50Indices0.410.30+47%-37.1%54%71
US 30YBonds0.070.18+11%-18.1%37%71
Hang SengIndices0.120.17+17%-44.6%51%69

Buy & hold = passive benchmark · Strategy = V3 engine, costs included · % positive windows = share of winning 42-day windows

The portfolio is the product

A Sharpe of 0.8–1.0 on every single asset does not exist — not for us, not for anyone. Published systematic strategy indices from major banks top out at 0.3–0.5 per asset class. The real edge comes from combining decorrelated signals: our equal-weight portfolio of all core signals reaches a Sharpe of 1.36 with a −10.4% maximum drawdown, versus −31% for Bitcoin alone. That is exactly how professional CTAs build performance.

1.36
Portfolio Sharpe
-10.4%
Max drawdown
71%
Positive 42-day windows
2014–2026
Measured period

What we tested and rejected

Being transparent about failures matters more than showing winners. Each of these was implemented, measured on real history, and thrown away because it degraded performance:

Perpetual funding rates

7 years of hourly data (2019–2026). Cutting exposure on hot funding: BTC 0.86 → 0.72. High funding actually precedes the best forward returns.

Crypto Fear & Greed index

Cutting exposure on extreme greed: BTC 0.86 → 0.56. The popular contrarian intuition is empirically false at a 5-day horizon.

FX carry with real central bank rates

8 central banks (Fed, ECB, BoJ, BoE, SNB, RBA, BoC, RBNZ). Cross-sectional carry basket: 0.26 Sharpe. Not tradable.

LLM directional vote (V2)

Removed in V3. An LLM opinion on direction cannot be backtested without look-ahead contamination, so it never enters the signal.

Where the model does not work — and why we tell you

Trend-following does not work everywhere. On FX we ran 40+ backtests across 10 families — mean reversion, breakout, cross-sectional momentum, and carry using real policy rates from 8 central banks — and nothing was tradable. Same on silver, copper and XRP, where no variant beats simply holding the asset. So we issue no directional signal on them: they show as NEUTRAL with macro context only. Hiding this would be dishonest.

USD/JPY+0.09
EUR/USD+0.05
AUD/USD+0.05
GBP/USD-0.05
USD/CAD-0.17
EUR/JPY-0.17
NZD/USD-0.27
EUR/GBP-0.52
USD/CHF-0.56
GBP/JPY+0.15
Silver+0.24
Copper+0.32
XRP+0.49

Assets above are excluded from directional signals. We still provide their institutional positioning and macro reading — just not a trend signal we cannot stand behind.

How it was computed

Fixed momentum horizons — 21 / 63 / 126 / 252 days — chosen before testing and never optimized on the data.

Volatility targeting: position size = target volatility ÷ realized 60-day volatility, capped at 1.5×, then smoothed over 20 days.

42-day rolling windows across the whole history; Sharpe, drawdown and share of positive windows computed on that stream.

Transaction costs applied on every position change (10 bps crypto, 5 bps indices and commodities, 3 bps bonds).

No LLM anywhere in the direction: the AI writes the explanation, it never votes on the signal.

Past performance, including backtested results, does not guarantee future returns. These figures describe the deterministic quant engine and are provided for transparency, not as investment advice.