Delta-neutral strategy
The delta-neutral strategy holds a long leg and an equal and opposite short leg in the same asset, so the two offset and the position carries almost no net price exposure. The return comes from carry rather than from price direction: one leg pays the other while the price move of either leg cancels.
Simulation only. This strategy runs against the deterministic backtest harness in @meridian/strategies. It is not wired to a live venue, a vault, or mainnet funds. The hedge venue is modeled inside the harness: there is no live perpetual or dated-futures venue, no order router, and no key that can trade. Every number this page describes is produced by the simulation from a scenario, not observed from a live account.
How the position is built
A delta-neutral position starts from a capital amount and a target leverage. The strategy holds the long leg in the base asset and shorts the hedge leg for the same notional, so the two deltas cancel:
| Quantity | Definition | Note |
|---|---|---|
| Long notional | capital × targetLeverage | Long spot leg |
| Short notional | Equal to the long notional | Hedge leg, sized 1:1 |
| Margin per leg | long notional ÷ targetLeverage (equals capital) | Posted on each leg |
| Total margin | 2 × capital | Both legs share the target |
Because the hedge is sized 1:1, the net delta of a freshly opened position is zero. The position earns the spread between the two legs, not the price move of either one.
Target leverage
targetLeverage is the gross leverage applied to each leg, between 1 and 20 inclusive. A value of 1 is an unlevered position; the upper bound of 20 keeps the position inside typical exchange initial-margin limits. A value outside the range, or a non-positive capital, is a typed error and the position is not opened.
Higher leverage scales both notional and margin proportionally. It does not change the direction of the trade, but it does shorten the distance to a liquidation of either leg, which is the risk the rebalance band and the liquidation path exist to manage.
Rebalancing to neutral
The two legs are only neutral at the moment they are opened. As prices move, the legs drift apart and a net delta reappears. The strategy measures the net delta in basis points of total equity and rebalances when it leaves the configured band:
netDeltaBpsisnet delta ÷ total equity × 10000, where the net delta is the spot value minus the short perp value.neutralityBandBpsis the allowed drift, in basis points.- When
|netDeltaBps|exceedsneutralityBandBps, the strategy trades the perp leg back to match the spot notional.
The trigger and the size of the correcting order are pure functions of the position state, so the same state always produces the same decision. That is what makes a backtest reproducible: a scenario plus a seed replays to the same trades.
A wide band trades less and tolerates more drift; a narrow band keeps the position closer to neutral but pays more in fills and fees. The band is a scenario input, and the trade-off is visible in the run summary as totalRebalances against maxNetDeltaBps.
Carry sources
Delta-neutral return is carry: the payments one leg makes to the other plus the yield on the collateral. Three sources are modeled.
- Funding. A perpetual hedge pays or receives a periodic funding rate. A positive rate is paid by longs and received by shorts; a negative rate reverses that. The strategy's short hedge therefore receives positive funding and pays negative funding. Funding accrues over each interval aligned to the simulation clock, and a gap in the series accrues zero rather than interpolating.
- Basis. When the hedge is a dated future rather than a perpetual, the gap between the spot price and the future price converges as the future approaches expiry. A short future opened above spot earns that convergence.
- Collateral and borrow spread. The long base leg can earn a supply rate while the leveraged legs pay borrow interest on their margin. The net of the two is part of carry and is accrued each step.
Carry is the sum of the three and can be negative. The strategy is market-neutral in price, not in carry: a negative funding regime, or a basis that moves against the hedge, loses money even when the net delta is zero.
Configuration
A backtest is driven by a scenario plus the strategy configuration. Money and rate values are fixed-point decimal strings, never JavaScript numbers, and carry at most seven decimal places. The strategy configuration (DeltaNeutralConfig) exposes:
| Option | Meaning | Safe range / Type |
|---|---|---|
asset | Base asset held long on spot | Valid asset symbol (e.g. XLM) |
targetLeverage | Gross leverage applied to each leg | 1 to 20 inclusive |
rebalanceBandBps | Rebalance threshold in basis points of total equity | Positive; commonly 20 to 50 |
fundingRate | Per-second funding rate applied to the short hedge | FundingRate instance |
priceFeed | Spot marks feed for the long leg and hedge venue | PriceFeed instance |
hedgeFeed | Optional perp marks for hedge (defaults to priceFeed) | PriceFeed instance (optional) |
hedgeVenue | Optional modeled hedge execution venue | HedgeVenue (optional) |
id | Optional stable identifier recorded in the snapshot | String (optional) |
Scenario-level parameters (e.g. startingCapital, time steps, and seeds) configure the overall backtest runner environment rather than the strategy itself.
The scenario that carries them is the single input a run is reproduced from: the same scenario and seed always yield the same run. A scenario describes a time window and step, the traded assets, the price and rate source, the starting capital, the strategy block, and the seed, plus a series of steps holding the spot price, the perp price, and the funding rate.
Running a scenario backtest
Scenarios are generated deterministically from a seed. The package ships generators for three market regimes:
- a trending market with a persistent basis and steady positive funding,
- a ranging market that mean-reverts,
- a high-funding regime that stresses carry.
Each generator takes a seed and a step count, so a scenario is reproducible on any machine. Run the strategy package, which validates the scenario and drives the backtest runner and its tests:
pnpm --filter @meridian/strategies test
pnpm --filter @meridian/strategies typecheckThe runner advances the clock one fixed step at a time. On each step it accrues funding and interest, asks the strategy for a rebalance action, applies any trade with slippage and fees, and records a snapshot. Every amount is bigint fixed-point, so identical inputs produce byte-identical runs. When the run ends, the metrics collector summarizes it.
Reading the run summary
The summary is the output to read. The runner outputs a RunReport object exposing standard risk metrics, discrete event counters, and portfolio state:
| Metric / Field | Meaning | Status |
|---|---|---|
riskMetrics.maxDrawdown | Maximum drawdown from peak portfolio value as a ratio (0 to 1) | On RunReport |
riskMetrics.sharpeProxy | Sharpe-ratio proxy: mean excess return in stroops per step | On RunReport |
riskMetrics.volatility | Standard deviation of per-step returns | On RunReport |
eventCounts.rebalances | How many times the strategy traded back to neutral | On RunReport |
finalState.totalValue | Final total portfolio valuation in reference asset (USDC) | On RunReport |
netPnl (planned) | Total change in equity over the window | Planned |
carryPnl (planned) | Equity earned from funding and carry | Planned |
pricePnl (planned) | Equity from price moves, which should stay small when neutral | Planned |
maxNetDeltaBps (planned) | Largest net-delta drift seen, in basis points | Planned |
withinNeutralityBand (planned) | Whether every step stayed inside rebalanceBandBps | Planned |
A healthy run keeps net delta inside rebalanceBandBps, minimizes drawdown, and limits excessive rebalance counts so trading costs do not erode returns. Attribution metrics like carryPnl and pricePnl are planned analytical extensions to break down the components of total PnL.
Risks and limits
- The hedge venue is modeled. The simulation assumes the hedge leg fills at the modeled price with the modeled cost and slippage. It does not model exchange outages, delisting, or a funding regime change a live venue could impose.
- Carry can invert. Funding and basis are not stable. A position that is neutral in price can still lose money when carry turns negative.
- Leverage can liquidate. A fast adverse move can push a leg through its liquidation threshold before the next rebalance. This is a real risk, not an edge case.
- Rebalancing is not free. Every correcting order pays a fill and a fee. A band that is too narrow costs more than it protects.
- Simulation only. No part of this strategy touches user funds. It exists to answer, deterministically, whether a given scenario earns carry worth the risk.
Related
packages/strategiesholds the SDK: fixed-point money math, price feeds, position sizing, the delta-neutral strategy, the scenario generators, the backtest runner, and the metrics collector.