Methodology notes — Delta-Neutral Carry
GLACIER is our delta-neutral carry strategy, currently in real-time paper validation: long spot against a short perpetual of equal notional on a single venue, capturing funding and basis while staying continuously delta-hedged.
That sentence is accurate, and taken alone it is quietly misleading. "Delta-neutral" describes one risk that has been removed. It says nothing about the ones that remain — and the ones that remain are what actually ends carry books. This note is about how we enumerate them, and what the design does in response.
Neutral to price is not neutral to ruin
A hedged carry position has a characteristic payoff shape: many small, quiet accruals punctuated by rare, fast losses. That shape does two things to a research process.
First, it makes the strategy look better the shorter you observe it. Calm is the mode, not the exception, so a short window is a flattering window by construction.
Second, it moves nearly all of the interesting risk into the tail — which means effort spent optimizing the average outcome is effort spent on the least important part of the distribution.
So we invert the usual order of work. Before asking how much carry a configuration harvests, we ask what specific sequence of events takes it to zero, and whether that sequence is survivable by construction rather than by luck.
The failure modes we plan around
The liquidation path
The short perpetual leg is the fragile one. In a sharp spot rally, the hedge — the leg doing exactly its job — is also the leg burning margin. The spot gain is unrealized, and possibly unusable as collateral in the same minute the perp leg needs it.
The dangerous scenario is not "the market went against us"; in aggregate, delta-neutral means it can't. It is "the market moved fast enough that our margin arrived later than the exchange's deadline." Liquidation avoidance is therefore a first-class design constraint, not a metric we inspect after the fact.
Carry that turns into a bill
Funding and basis are regimes, not constants. They can compress toward nothing, and they can invert and stay inverted long enough that a position which is "working" costs money at every settlement. A carry strategy with no defined exit for a regime change isn't a carry strategy — it's a bet that the last regime continues.
The venue is a position
Running both legs on one venue removes the worst execution problem: collateral stranded on the wrong exchange when you need it in minutes. It replaces that with concentration — outage, withdrawal halt, auto-deleveraging, a margin-rule change mid-stress.
We prefer a risk we can size and monitor over one that only appears during the exact event we hedged for. But that is a trade, not a free lunch, so venue exposure is treated as an explicit position with a cap, never as an assumption baked into the model.
The seam between two legs
A delta-neutral book is only neutral once both legs exist. Entry and unwind are the moments the strategy is directionally exposed, and they tend to arrive when liquidity is worst. Slippage, partial fills and one-sided fills are execution risks that no amount of signal quality will fix — which is precisely why forward validation exists.
What the design does about it
Buffers sized against stress history
Sizing to recent volatility gives a carry book exactly enough margin for the conditions that just ended. Our validation spans multi-year windows that include major dislocations — the LUNA collapse, the FTX failure — so the worst case in the parameterization comes from the record rather than from imagination.
Stand-down rules written first
We define what makes us not trade before we define what makes us trade: degraded depth, data-integrity gaps, a carry level that no longer compensates for the tail being underwritten. A rule written in advance is a rule. A rule written during the event is a rationalization.
Paper-forward as the gate
The failure modes above are execution-shaped, not signal-shaped, so no backtest can clear them. GLACIER is in real-time paper validation today. It graduates when the operational surface — fills, latency, margin mechanics, reconciliation, alerting — has been observed under live conditions, not when the research looks convincing.
What we don't do
- We don't lever a thin spread to make it interesting. Leverage converts a slow positive-skew grind into a fast negative-skew one.
- We don't count unrealized basis as earned.
- We don't split legs across venues we cannot rebalance between quickly.
- We don't let an automated system re-enter after an unexplained loss before a human has reviewed it.
- We don't publish performance figures. We publish the process, never the performance numbers.
A note on the current tape
As of this writing, BTC and ETH have each moved less than half a percent over 24 hours, the broader large-cap perpetuals are mixed, and the sentiment index sits in Fear territory in the mid-40s.
A quiet tape with cautious sentiment is the regime in which carry looks easiest and buffers feel expensive. It is also the regime in which sizing decisions actually get made — which is why we answer the sizing question against the stress record instead of the calm one.
GLACIER is in paper validation. We do not manage client assets. Strategy overviews and current stages: crynomad.ai/strategies. Performance data is shared individually with qualified investors on request.