Problem: liquidate inventory Q under temporary (k) and permanent (b) impact, running funding β(P−ψ(S)), inventory penalty φ and terminal penalty α(Q_T)^2 .
Main result: for ψ(s)=s the optimal trading speed is linear feedback
ν*(t,q,p,s) = (1/4k)[ (ξ(t)+π(t)) q + (1/b)(ξ(t)−π(t))(p−s) ], with closed-form ξ, π given by Riccati-ODE solutions (Theorem 2) .
Economic interpretation: coefficients on q and on funding-rate proxy z=p−s are negative (agent accelerates liquidation to reduce inventory risk and to manage funding payments) .
Small-β (funding rate) asymptotic: Almgren–Chriss base + β-correction
Constructed approximate control:
ν̂(t,q,p,s;β) = ν0(t,q) + β ν1(t,q,p,s),
where ν0(t,q) = (1/2k)(b + 2γ(t)) q (Almgren–Chriss-like) and ν1 contains an "alpha-like" signal term proportional to the expected funding signal E[p−ψ(S_u)|S_t=s] integrated over remaining horizon .
Practical takeaway: use ν0 for baseline execution; add β-correction to exploit expected funding-rate advantage while accounting for additional inventory risk .
Short-time (small T) asymptotic and closed-form substitution rule
Short-time expansion: hψ(t;T) = h̃0(q) + (T−t) h̃1,ψ + (T−t)^2 h̃2,ψ + o(T^2), with h̃0 = −α q^2 and explicit first-order correction capturing immediate funding impact and inventory penalties .
Resulting approximate control (truncate linear-in-(T−t) terms):
ν̃(t,q,p,s;T) = ν̃0(q) + (T−t) ν̃1(q,p,s),
where ν̃0 = −(2α−b)/(2k) q and ν̃1 contains −(β/2k)(p−ψ(s)) plus inventory-penalty adjustments .
Closed-form approximation: substituting ψ(s) for the spot argument in the identity-payoff closed-form strategy (ν*(·,·,·,ψ(s))) yields an admissible control that is asymptotically optimal to O(T^2) for small T (Proposition 9) .
Strategy comparison, simulations and practical guidance
Numerical experiments compare: Almgren–Chriss (ignore funding), short-time ν̃, and closed-form-substitution ν̄ (use identity-ψ strategy with ψ(s) plugged in). Payoff examples: logistic and quadratic perturbations of identity (Fig.4) .
Findings:
As T→0 all strategies converge to −α Q0^2 (terminal penalty dominated) .
For small but non-negligible T, ν̄ (closed-form-substitution) typically outperforms ν̃ because it leverages the global- T solution structure from the identity case; ν̃ can degrade as T grows .
Simple decision rule:
If ψ≈identity or T is very small, use ν̄ (plug ψ into closed-form identity strategy) .
If β is small and longer horizons, use ν̂ (Almgren–Chriss base + β-correction) to capture expected funding alpha while controlling inventory .
Monitor temporary impact parameter k: as k→0 agent should maintain A_t ≈ 0 where A_t=(bβ+2φ)Q_t+βZ_t, per Proposition 3 (rule-of-thumb) .
思维导图
深度阅读
元数据与概览(引言与报告概览)
报告标题:Optimal Liquidation of Perpetual Contracts,
作者与机构:Ryan Donnelly、Junhan Lin(King’s College London),Matthew Lorig(University of Washington)。
解读:左面板(初始 P higher)显示更快初期清算(比 AC 更快),中间面板(初始 funding=0)在早期近似 AC 路径但中后期持仓更高随后在末端加速清算,右面板(初始 P lower)则相对滞后清算——这些行为可归因于交易对 P 的永久冲击改变资金费率,从而改变持仓收益/成本的即时激励。