The long sleeve playbook

The long side of inite.fund is a four-asset portfolio that rebalances weekly. Composition is fixed; weights float. The decision tree behind the weights is the whole story.

The four sleeves

• Crypto — BTC + ETH spot. Risk asset. Carries beta to global liquidity, plus its own cycle.
• Equity — SPY (or its IBKR equivalent for non-US users). Risk asset with a different driver (earnings + rates), low-correlation tail to crypto in calm regimes, high-correlation in panic.
• Gold — GLD. Anti-correlation to real rates, decent during USD-debasement narratives, useless in pure deflation.
• Bonds — SHY (1-3y Treasuries). Cash equivalent in low-rate world; non-zero hedge in deflation. We don’t run long duration — that’s a separate trade.

Why these four and not eight? Because adding sleeves doesn’t add information past a point — it adds operational overhead. With four we can rebalance in a single rebalance event, fit the entire allocation in one venue (or two: crypto on Binance, the rest on IBKR), and explain the regime call to a human in three sentences.

The composite

Each sleeve has a composite score in [-1, +1]. The composite is a weighted blend of:

• Trend — multi-timeframe moving averages, breakout proximity.
• Breadth — for equity, % of constituents above 200d. For crypto, % of top-50 above their 50d.
• Regime tags — HMM-detected state (bull / chop / bear) on the sleeve’s own returns. The classifier itself — observation vector, three-state choice, calibration window — is in the regime engine, in one note.
• Cross-asset — correlation to the others, dollar strength, real rates.

The composite isn’t designed to time tops or bottoms. It’s designed to answer “would a disciplined person hold more or less of this right now”. That’s all. We don’t run alpha factors. We don’t trade headlines.

From composite to weight

We map composite → target allocation through a piecewise-linear function calibrated per sleeve. Crypto can go from 5% to 35% based on its composite. Equity from 10% to 45%. Gold 5–15%. Bonds whatever’s left, which is also the cash buffer.

Two important constraints:

1. No sleeve goes to zero. Even at composite = -1, the floor is ~5%. Reason: regime detectors are wrong sometimes, and being completely out of an asset means you can’t catch the early-recovery move that is, by definition, hard to detect.

2. Sum of risk sleeves capped. Crypto + equity + gold ≤ 80%. Forces at least a 20% bond/cash position even when everything looks green. We’ve never regretted this.

The drawdown governor

Composite is a function of market state. The governor is a function of your state. If portfolio drawdown crosses -8%, the governor halves new allocation deltas — not the existing positions, just future moves. Past -12% it freezes new capital deployment entirely until DD recovers.

This is not a stop-loss. We don’t dump positions. Selling at -12% locks in losses; freezing new deployment just slows reallocation. When the governor is active you’ll see dd_governor_active: true on the strategy state — that’s a signal to humans, not to the engine. Operators can override. The math behind the throttle (why -8%, why continuous, why three earlier designs broke) is in drawdown as a throttle, not a brake.

The reserve pool

When equity grows past starting capital, a fraction skims into a reserve pool. The pool sits in bonds and waits. When the composite shows oversold reads on a risk asset (composite < -0.5) AND the asset’s position is at floor, the pool deploys.

This is anti-cyclical by construction. You’re banking gains during trending bulls and re-deploying during panics. We don’t need to time the panic perfectly — we just need a non-zero amount of dry powder ready when composite crosses the threshold.

The reserve pool is the single most underrated part of the strategy. In backtests it added ~3pp/yr without changing the rule set anywhere else.

Why weekly

We rebalance once a week. Wednesday night UTC, by default. Why not daily? Three reasons:

• Costs. Crypto fees + slippage at 4 venues, equity commissions, spread on GLD/SHY — daily rebalancing eats 30–60bps/yr of return for no decision-quality improvement.
• Signal vs. noise. Composites at daily resolution are noisy. The signal we care about is “regime is shifting” — that’s a multi-week thing.
• Operator load. Weekly is a sane cadence for a human to look at the rebalance preview, approve, and move on.

Where it isn’t great

The portfolio underperforms in two specific environments:

1. Sustained low-vol melt-ups. When everything goes up monotonically, our cap on risk sleeves leaves money on the table vs. a 100% SPY benchmark. We accept this — the hedge against the alternative (sustained low-vol crashes) is worth more than the giveaway.
2. Regime whipsaws. Late 2018, March 2020, early 2022 — the regime detector flipped 3+ times in 6 weeks. Each flip costs turnover. The drawdown governor helps but doesn’t eliminate it.

We monitor turnover-per-quarter as a sanity check. Above 25% per quarter, the regime detector is over-fitting to noise — we re-tune.

The honest pitch

This is not a high-Sharpe momentum machine. It’s a hedge-fund-shaped allocation policy that any disciplined person could run by hand if they had a couple hours a week and the muscle to ignore CNBC. The software’s value is consistency, audit trail, and the integration with our trading sleeve (so you can park 80% in this and run a 20% intraday trading allocation on the same balance).

If you want 10x leverage in an idea you read on Twitter, this isn’t it. If you want a sleeve that grinds out 8–14%/yr with -10% to -15% max drawdown across full cycles, this is the shape of it.

The live track record is on /strategies.