Skip to content
log(swap)

Fees as a path statistic

The fee is a rate on displacement

Charge a percentage ϕ\phi of the quote leg. By I2 the quote leg is linear in xx, so

fee=ϕ L ∣dx∣⟹dF=ϕ ∣dx∣,F=ϕ⋅TV(log⁡p),\text{fee} = \phi\,L\,\lvert dx\rvert \quad\Longrightarrow\quad dF = \phi\,\lvert dx\rvert, \qquad F = \phi\cdot\mathrm{TV}(\log p),

with FF measured per unit of exposure. The pool's fee state is a geometric property of the price path — ϕ\phi times its total variation — with no price-weighted Jacobian to reconcile.

A base-side fee does not take this form: it carries a 1/p1/p weight, so the accumulator would no longer be ϕ TV(log⁡p)\phi\,\mathrm{TV}(\log p) and one scalar would not serve every position. That is why fees here are quote-only, and it is a consequence rather than a convention.

FF is also the yield

Because FF is per unit LL, and LL is the base-leg value in quote, FF is directly a cumulative return on base exposure — no normalisation:

feesi=Li ΔF,feesiLi=ΔF,feesiVi=ΔF1+ℓi.\text{fees}_i = L_i\,\Delta F, \qquad \frac{\text{fees}_i}{L_i} = \Delta F, \qquad \frac{\text{fees}_i}{V_i} = \frac{\Delta F}{1 + \ell_i}.

So ΔF\Delta F over a window is cumulative fee return on active liquidity. Reporting an APR requires annualizing by the window duration. For an active uncapped position, dividing by its starting (1+ℓ)(1+\ell) converts it to return on starting capital — the same (1+ℓ)−1(1+\ell)^{-1} that appears as leverage. One global scalar answers "what is this pool yielding" for every LP at once. A price-weighted accumulator would not survive this.

The edge

Track the quadratic variation Σ=∫(dx)2\Sigma = \int (dx)^2 alongside FF. Realized LVR per unit LL is exactly 12Σ\tfrac12\Sigma on the path that happened — the same 12σ2L\tfrac12\sigma^2 L with realized variance in place of an assumed one. Hence

edge per unit L  =  F−12Σ.\text{edge per unit } L \;=\; F - \tfrac12\Sigma .

Two accumulated numbers and a subtraction. No volatility model, no benchmark portfolio, no oracle. They are commensurable because both are path statistics of the same log⁡p\log p.

Simulation check. σ=60%\sigma = 60\%, T=1T = 1 year, Δt=12\Delta t = 12s, 40 paths: E[F]=0.180013\mathbb{E}[F] = 0.180013 against E[12Σ]=0.180020\mathbb{E}[\tfrac12\Sigma] = 0.180020, theory 12σ2T=0.18\tfrac12\sigma^2T = 0.18, with per-path edge within ±10−4\pm 10^{-4}.

Calibration: the matched gain

Set ϕ=κσ\phi = \kappa\sigma and equate expected income with the bill. Per block, fee income is ϕL E∣Δx∣\phi L\,\mathbb{E}\lvert\Delta x\rvert and the LVR bill is 12σ2LΔt\tfrac12\sigma^2 L\Delta t. For Gaussian log-returns E∣Δx∣=σ2Δt/π\mathbb{E}\lvert\Delta x\rvert = \sigma\sqrt{2\Delta t/\pi}, so

ϕ⋆(σ)=σπ Δt8,κ⋆=π/8≈0.6267.\phi^\star(\sigma) = \sigma\sqrt{\frac{\pi\,\Delta t}{8}}, \qquad \kappa^\star = \sqrt{\pi/8} \approx 0.6267 .

ϕ⋆\phi^\star is linear in σ\sigma — that is the central calibration fact. LVR is linear in variance while fee revenue is linear in volatility, so a single static ϕ\phi under-compensates LPs in volatile regimes and over-charges traders in quiet ones. The two errors point in opposite directions and no static value fixes both.

Worked example. σ=50%\sigma = 50\% per year, Δt=12\Delta t = 12s ≈3.80×10−7\approx 3.80\times10^{-7} yr: ϕ⋆≈1.93×10−4≈2\phi^\star \approx 1.93\times10^{-4} \approx 2 bp per swap.

At κ=κ⋆\kappa = \kappa^\star the edge is zero in expectation by construction, so the sign of the realized F−12ΣF - \tfrac12\Sigma reports directly whether the kernel over- or under-charged on the path that happened. The current contracts charge fixed swap fees. Fee machinery separates that implementation from the earlier dynamic estimator.

Fees translate the floor

A quote injection ff at fixed (L,x)(L, x) cannot leave the position unchanged. Since Q=L(x−ξ)Q = L(x - \xi), it is a floor translation:

Δξ=− f/L.\Delta\xi = -\,f/L .

Check: R eQnew/L=(L/p)(p/pmin⁡)ef/L=L/pmin⁡′R\,e^{Q^{\text{new}}/L} = (L/p)(p/p_{\min})e^{f/L} = L/p'_{\min}. ✓

Compounding is therefore a relabelling rather than a trade — there is no ratio to rebalance into, and no swap is needed. Quote-only fees are precisely what makes the identity work; base-asset fees would require a market swap back to quote first.

Two things follow, one for each pool:

  • In the F-pool this identity is the entire fee mechanism: every share has the same floor, so moving it credits everyone at once and nothing is stored.
  • In the C-pool it is the same additive shift for every active position, independent of LiL_i and ξi\xi_i, so ladder spacing and ordering are preserved by compounding.

The dual direction is also available: withdrawing qq of quote raises the floor by q/Lq/L, bounded by q≤L(x−ξ)=Qq \le L(x - \xi) = Q — drainable to the current price, not beyond.