FUTURE3.AI
Systematic Strategies · Multi-Strategy · Rev. 2026
Systematic Multi-Strategy Fund

FUTURE3.ai

Four independent engines. Two markets. One discipline.

Across digital assets and China A-shares — from market-neutral carry and liquidity provision to a disciplined, long-only trend system. Each strategy stands on its own independent test; the edge is the portfolio.

Strategies
04
Independent · low mutual correlation
Markets
02
Digital assets + China A-shares
Directional beta
~0
Neutral sleeves ≈ zero to BTC
Systematic
100%
Rules-based · all post-cost

Net-value curves below are indexed to 1.0 and post-cost (trading costs and real funding rates included). The vertical axis shows net value as a multiple of the 1.0 start (1×, 2× …); we still don't quote Sharpe or annualised return. See disclosures · June 2026.

01 / The Idea

One discipline, four uncorrelated return streams

We don't bet on a single signal. Every strategy must independently survive look-ahead-free testing and still stand after real costs and funding — only then does it join the book.

A market-neutral carry sleeve and a liquidity-provision overlay grind out low-volatility return regardless of direction. A cross-sectional equity engine harvests relative value in China A-shares. A long-only trend system captures crypto's upside while sitting in cash through its worst drawdowns. Different markets, different mechanisms, low mutual correlation.

02 / The Strategies

Four engines

Each cleared its own independent test. The readouts are net value, indexed to 1.0 — the vertical axis shows the multiple of the 1.0 start.

01 / FUNDING CARRY
Digital assetsMarket-neutralUnlevered

Market-Neutral Funding CarryLong spot, short equal-notional perp — earn the rate spread.

Buy spot and short an equal-notional perpetual. The two legs offset on price; the strategy earns only the funding-rate spread of the perp over spot. Direction doesn't drive P&L.

Run inside a Binance portfolio-margin account — spot long + perp short, never levered, no naked alt exposure. The core mechanism is adaptive deployment: sit in cash when the edge is thin; it still printed positive through the February 2026 funding trough.

Full report →
1.1×1.2×20232026
Net value · indexed 1.0 · costs & funding in · ×multiple of 1.0
02 / MARKET-MAKING
Digital assetsMarket-neutralExecution edge

Liquidity & Market-Making OverlayPassive quoting on wide-spread, low-alpha names.

Provide liquidity passively where spreads are wide and alpha is thin, capturing spread and rebate; switch to taking when spreads tighten or signal appears. No directional exposure.

A two-leg cross-venue maker book (spot quote, perp hedge), with a fill model corrected for look-ahead and calibrated on Binance full-pool trade data. It is a capacity-constrained measuring instrument — high quality, low capacity — positioned as an execution-timing overlay, not a scalable return centre.

Presented honestly as an execution edge — not a headline-return product.

Mechanism
Passive quotingSpread + rebate capture
Risk
Market-neutralZero directional exposure
Capacity
Low · per-nameDeliberately bounded
Role
Execution overlayTiming layer for other sleeves
◐ No standalone P&L shown
03 / A-SHARE SELECTION
China A-sharesCross-sectionalRelative value

Cross-Sectional A-Share SelectionDaily factor ranking across the CSI 300 universe.

Score and rank CSI 300 constituents daily on a robust set of cross-sectional factors, hold the top basket, and earn excess return over the benchmark.

15 features · LSTM factor mining · multi-seed robustification, validated on held-out data. Tracking through June 2026 kept beating the benchmark and stayed positive through a falling market — genuine stock-picking; current names cluster in semiconductors, components and communications.

1.5×20242026
Excess vs benchmark · indexed 1.0 · purged WF · ×multiple of 1.0
04 / BITCOIN TREND
Digital assetsLong-only or cashNever short · never levered

Bitcoin Trend TimingHold the upside; sit in cash through deep drawdowns.

Hold bitcoin's upside on a systematic dual signal and step to cash to avoid deep drawdowns — long or cash only, never short, never levered.

Target weight = 50% trend signal + 50% machine-learning signal ∈ {0%, 50%, 100%}. Execution timing is phase-tested. On average 42% of capital sits in cash, fully flat on 21% of trading days — comparable total return, roughly half the drawdown.

Full report →
20222026
Net value · indexed 1.0 · flat segments = cash · ×multiple of 1.0
03 / Diversification

Spread across the neutrality spectrum

The four engines sit at different points on the market-neutral → directional axis — which is where the diversification comes from. Two neutral sleeves form a low-volatility base; A-shares and BTC supply directional return. Marker size shows relative risk-adjusted quality; colour shows market.

Market-neutral Directional Low vol High return · high vol 02 · Market-making no standalone P&L 01 · Funding carry market-neutral 03 · A-share selection relative value 04 · Bitcoin trend long or cash
Digital assets · neutral China A-shares Digital assets · directional ○ Size ∝ relative quality · dashed = overlay
04 / At a Glance

The book, in one table

No specific return figures — what's compared here is structure, not performance.

Post-cost · structure comparison only.
StrategyMarketStyleExposurePeriod
Funding CarryRate spread Digital assetsCarryMarket-neutral2023–2026
Market-MakingExecution edge Digital assetsLiquidityMarket-neutral
A-Share SelectionCross-sectional China A-sharesRelative valueLow · vs benchmark2024–2026
Bitcoin TrendTrend timing Digital assetsTrendDirectional · long/cash2022–2026
05 / How We Build

The honesty is the moat

01

No look-ahead

Every curve uses only information available before each decision date, with an embargo between training and evaluation. Execution timing is phase-tested so no future information aligns a fill.

02

Costs always on

Per-trade costs, per-name borrow and maker spreads, and real funding rates booked tick-by-tick — all included. We report post-cost behaviour, not the illusion of gross.

03

We publish the failures

Reinforcement-learning timing, momentum mining, higher-frequency variants — many attempts were killed by our own adversarial review. What survives is the few that withstand counter-evidence.

What we don't claim