humarb ResearchMethodological preprint · 002 · Zenodo v1

Framing Net Carry for Cross-Venue Perpetual Funding Positions

A Settled-History, Depth-Walked, and Liquidation-Aware Method

Download paper Read the abstract
Perpetual futuresFunding ratesNet carryExecution costsLiquidation riskForecast evaluation
Abstract95 words

A net-carry number should expose its accounting, timing, costs, and omissions.

Cross-venue perpetual funding positions are often summarized by annualizing the current funding-rate spread, a practice that conflates transient quotes with holding-period cash and omits execution and margin risks. This paper develops an accounting framework based on settled funding history, asynchronous settlement grids, depth-walked entry and exit costs, and tier-adjusted liquidation distances. Net income is reported at reference and conservative ends, with unavailable inputs left explicitly unavailable. A nominal cost-coverage horizon is distinguished from recovery under time-varying funding. A limited replay motivates the historical basis but does not validate the estimator; out-of-sample performance is not reported.

阅读中文摘要

跨交易场所永续合约资金费套利常以“当前费率差”年化展示，但该指标把瞬时报价误作持有期现金流，并忽略不同结算时刻、交易规模对应的盘口成本、出场成本及单腿强平风险。本文提出一个可审计的净持有收益会计框架：以已结算资金费历史的稳健汇总形成参考基准，在单次结算层面应用费率上下限，并在混合结算网格上计算两腿现金流；以四次吃单成交、逐档盘口成交成本、已测量的其他成本和明示的价差回归项估计净收入。结果分为参考端与保守端，后者是压力情形而非置信区间。关键输入缺失时，估计不可用而非以默认值补齐。本文还区分名义成本覆盖时长与时变资金费路径下的真实覆盖时间，并以维持保证金档位和近期方向性单日波动构造强平缓冲指标。有限的离线 replay 仅用于说明当前费率外推的偏差，不构成对生产估计器的样本外验证；本文未报告该估计器的样本外结果。框架旨在明确测量、压力测试与概率预测的边界，不构成投资建议。

04Original schematic figures
16Numbered equations
19Checked bibliography records
00Claimed out-of-sample results
Paper contents

I · Research question

The latest displayed rate is not a holding-period cash flow.

A perpetual future has no expiry. Its anchoring mechanism is a sequence of transfers between long and short positions, not convergence at maturity. When a matched route holds a long leg on one venue and a short leg on another, the difference between their funding rates can become a meaningful source of cash income.

The simplest screen subtracts two currently displayed rates and annualizes the difference. That operation is immediate and comparable, but it assumes that observed rates remain representative, both legs pay on compatible schedules, and execution prices do not depend on trade size. This paper instead begins with the accounting identity that connects funding cash, execution costs, other measured frictions, and price-spread changes over a stated holding period.

The companion snapshot study asks what remains after measured costs. This paper asks how that number is framed—and what must remain visibly unknown.

Read Research 01: Measuring Executable Net Carry

II · Framework

One accounting identity. Two auditable ends.

The headline identity is deliberately additive. All cash amounts are signed on one long-leg/short-leg basis over a common holding period. The reference end is not the expected value of the conservative end; the two are accounting cases with different inputs.

Reference case
Πref = Gref − Kent − Kxit − Koth + Bref
I

Settled-history basis

A robust summary of officially settled funding prints replaces the latest displayed rate. Caps and dead zones are applied at the settlement scale rather than averaged away.

Cash timing
II

Exact settlement-block grid

Each venue’s interval boundaries are projected onto a common clock. One merged settlement block uses only events whose funding cash actually settles inside that block.

Mixed intervals
III

Depth-walked execution

Entry and exit are measured by walking displayed book levels at a stated common base quantity. Four taker fills replace the fiction that an entire position trades at the mid-price.

Size dependent
IV

Liquidation-aware cushion

Tier-adjusted isolated-margin distances are compared with each asset’s recent directional daily move. The nearer leg governs the route-level cushion.

Stress comparison

Reference and conservative are not confidence endpoints.

The conservative case combines less favorable funding, higher exit cost, adverse spread treatment, and an optional reserve. The outer minimum keeps it on the weak side of the reference case. No probability is attached to the interval between them.

III · Figures

The method, drawn without market observations.

01 / 04
Figure 1

Two legs settle on different clocks; their merged block has length Λ, and cumulative funding follows a staircase whose block-to-block change is d. Schematic only.

Figure 2

The conservative case combines less favorable funding, higher exit cost, and adverse spread treatment. The zero line is accounting breakeven, not a probability threshold.

Figure 3

The stepped VWAP path reflects consumption of successive book levels; the gap from a top-of-book approximation is size dependent. Schematic only.

Figure 4

The recent directional move is compared with each leg’s own approximate distance; the nearer comparison governs the route ratio. Schematic only.

Scroll horizontally or use the arrows. Select a figure to inspect its details.

IV · Evidence boundary

Measurement, stress, and probability remain separate.

The framework’s central discipline is semantic separation. A measured current cost is not a forecast fill; a conservative end is not a confidence bound; a nominal horizon is not capital recovery; and a historical cushion is not survival probability.

Implemented in the measurement framework

Settled-history construction, exact settlement accounting, depth-walked costs, freshness checks, nominal cost coverage, and liquidation-aware stress comparisons.

Planned evaluation methods

Proper scoring rules, interval tests, split-conformal calibration, headline outcome evaluation, PBO, CSCV, and deflated-Sharpe diagnostics are designs—not reported findings.

Limited historical replay

The replay motivates caution about current-rate extrapolation. It does not validate the estimator presented here or establish out-of-sample performance.

Nominal horizon · only when g > 0
T∞ = F + Kg hours

It answers a restricted flat-gap arithmetic question, not the time to recover capital under a time-varying funding path.

V · Limitations

An auditable conceptual framework, not a complete implementation specification.

  • No performance generalization. The paper reports no out-of-sample headline-estimator result, formal interval-coverage result, PBO, CSCV, or deflated-Sharpe result.
  • Current books are not future fills. A fresh book reduces stale-input risk but does not guarantee execution. Venues without an order book cannot receive a walked cost estimate.
  • Liquidation risk is approximate. The model covers isolated margin and a single position, uses single-day rather than cumulative movement, and excludes several venue-specific mechanisms.
  • Replication is intentionally bounded. Undisclosed production parameters and operating rules remain private; the contribution is the accounting and methodological frame.

This paper is for research and methodological information only. It does not provide investment, legal, accounting, or tax advice; does not recommend any trade or leverage; and does not promise income, execution, or protection from liquidation.

VI · Citation

Cite the method.
Preserve the evidence boundary.

Cite with the concept DOI to reference all versions. The Zenodo v1 record and its version-specific DOI remain available for version verification.

Concept DOI10.5281/zenodo.23897013
View the published v1 record on Zenodo

Liu, H. (2026). Framing net carry for cross-venue perpetual funding positions: A settled-history, depth-walked, and liquidation-aware method [Preprint]. Zenodo. https://doi.org/10.5281/zenodo.23897013

Download .bib Download PDF

Figure