Pith. sign in

REVIEW 2 major objections 4 minor 47 references

Cutting liquidity providers' share of Uniswap fees produced no large short-run drop in active liquidity or depth.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-10 05:58 UTC pith:NTNNJJEN

load-bearing objection Careful non-detection of short-run LP supply response to a real Uniswap take-rate cut, with strong reconstruction discipline and an honest identification boundary. the 2 major comments →

arxiv 2607.08525 v1 pith:NTNNJJEN submitted 2026-07-09 stat.AP

Causal Effects of Protocol-Fee Changes on Liquidity Provision in Automated Market Makers

Wen-Ting Wang This is my paper
classification stat.AP
keywords automated market makersprotocol feesliquidity provisiondifference-in-differenceson-chain dataUniswaptake-rateevent study
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper asks whether liquidity providers pull capital when a protocol takes a cut of their fee income, while traders still face the same posted fees. It uses the Uniswap protocol-fee switch of late 2025, which lowered the LP take-rate on many pools with tier-differentiated intensity, as a real policy shock. With a pre-specified matched-overlap event-study design and a frozen panel rebuilt from public logs before any estimate, it finds no large short-run average response in active liquidity or local depth, and more precisely estimated participation and composition outcomes likewise show none. The result is a non-detection at the design's resolution rather than a high-precision zero. The work matters because fee-controller simulations routinely freeze liquidity supply; this estimate bounds that frozen kernel on the LP side and cleanly separates it from trader-facing dynamic-fee claims the shock cannot identify.

Core claim

Under the frozen matched-overlap event-study design, the Uniswap protocol-fee switch that cut LP take-rates produces no large short-run average response in active liquidity or local depth among matched treated pools; LP participation and composition are more precisely estimated and also show no detectable response. The result is a non-detection at the design's resolution rather than a precise zero, and it survives the paper's pre-specified sensitivity checks. Token-1 volume and native fee income fail the parallel-trends gate and stay descriptive.

What carries the argument

The kernel K_L: the finite-horizon matched-overlap average treatment effect on the treated for LP-supply outcomes (active liquidity, depth, and related participation measures), recovered by a pre-specified event-study difference-in-differences design that moves only the LP take-rate while holding trader-facing fees fixed.

Load-bearing premise

After matching and baseline adjustment, treated and low-exposure control pools would have followed the same path in liquidity and depth if the fee switch had never happened.

What would settle it

Re-estimate the same event-study paths after regenerating the frozen panel with an independent reconstruction of active liquidity and depth; if post-period aggregate intervals for active liquidity and local depth then exclude zero and show a clear withdrawal, the non-detection claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 4 minor

Summary. The paper estimates the short-run causal effect of the Uniswap protocol-fee switch (executed 2025-12-28) on LP liquidity supply. The switch cut LP take-rates with tier-differentiated intensity while leaving trader-facing fees fixed, identifying an LP-side response kernel K_L rather than trader-facing dynamic-fee effects. Using a pre-specified matched-overlap event-study DiD design on a frozen, hash-checked panel reconstructed from public logs, the authors find no large average short-run response in active liquidity or local depth among matched treated pools; more precisely estimated LP participation and composition outcomes likewise show none. Token-1 volume and native fee income fail the parallel-trends gate and are reported only descriptively. A channel-admissibility audit delimits the estimand: K_L is design-based, while trader-facing dynamic-fee protection remains outside the empirical claim.

Significance. If the non-detection holds under the stated assumptions, the paper supplies a rare design-based constraint on the LP-exit margin that fee-controller simulations routinely freeze. The contribution is methodological as much as substantive: pre-specification, deterministic reconstruction of treatment/event time/outcomes into a hash-checked panel before any estimate, explicit parallel-trends gating with descriptive fallback, and a channel map that separates what the shock identifies from what it cannot. These practices raise the bar for causal claims from on-chain DeFi data and give a concrete empirical input (short-run K_L near zero at the design's resolution) for subsequent welfare and controller work. The result is carefully scoped as a non-detection rather than a precise zero, which is appropriate given the precision of the primary liquidity intervals.

major comments (2)
  1. A3 / Eq. (6) and Appendix A Figure 5: residual post-match SMD on log pre-period fee revenue remains 0.84 (from 1.19). Identification therefore rests almost entirely on dynamic conditional parallel trends after baseline-time adjustment rather than on level balance. The paper already reports this honestly and supplies Honest-DiD, placebo-date, entropy-balancing, window, and caliper checks that leave the null intact. Still, for the primary liquidity/depth outcomes whose aggregate CIs are wide (e.g., active liquidity [−1.11, 0.33] asinh), the manuscript should more explicitly quantify how large a differential untreated trend would need to be to overturn the non-detection reading, beyond the zero-breakdown Honest-DiD statement that follows from intervals already containing zero.
  2. Section 4.2 precision discussion: the design rules out only very large average responses for active liquidity (MDE near 1.03 asinh) and is still less precise for depth (MDE near 2.35). The participation nulls are well-powered. The central claim is already framed as non-detection at design resolution, which is correct; the abstract and conclusion should keep that language and avoid any implication that moderate short-run liquidity responses are ruled out. A short power/MDE table in the main text (currently only in Appendix A) would make the resolution transparent to readers who stop at the figures.
minor comments (4)
  1. Figure 2 identification-gate map is useful but dense; a one-sentence caption note explaining the joint pre-trend p-value threshold used for the colour coding would help.
  2. Notation: K_L is introduced as shorthand for the reduced-form LP-supply ATT path on M_A; a single explicit sentence in §3.1 equating it to the vector of Δ^m_h for m in M_A would prevent readers from treating it as a structural elasticity.
  3. References include several 2026 package citations (fixest, HonestDiD, ebal, fect); ensure arXiv/CRAN versions are stable or pin DOIs where available.
  4. Appendix B reconstruction pipeline is a strength; a one-paragraph main-text pointer to the subset bit-parity and estimator-parity checks would make the reproducibility claim more visible without expanding the main body.

Circularity Check

0 steps flagged

No circularity: empirical DiD non-detection under pre-specified design; estimand is reduced-form ATT path, not a fitted or self-defined prediction.

full rationale

This is an observational matched-overlap event-study DiD of a real Uniswap protocol-fee switch (2025-12-28). The primary object K_L is defined as the finite-horizon reduced-form ATT path of reconstructed LP-supply outcomes (active liquidity, depth, participation) for matched treated pools relative to low-exposure controls (Eq. 2, Section 3.1); the event-study coefficients of Eq. 3 target that path under dynamic conditional parallel trends (A3, Eq. 6) and the other stated assumptions. Nothing is fitted to a subset and then re-labeled a prediction; no uniqueness theorem or ansatz is imported from the author's prior work; the channel-admissibility audit routes trader-facing dynamic-fee claims outside the estimand rather than redefining them into the result. Reconstruction is deterministic from public logs into a frozen hash-checked panel before estimation (A1). The headline non-detection (wide post CIs that include zero; tighter nulls on participation) is an empirical measurement under those assumptions, not a quantity forced by construction. Residual post-match imbalance on the selection variable is acknowledged and does not create circularity. Score 0 is the correct honest finding.

Axiom & Free-Parameter Ledger

4 free parameters · 6 axioms · 1 invented entities

The central non-detection rests on standard DiD identification assumptions plus design choices (caliper, window, exposure restriction) and deterministic log reconstruction. No new physical entities are postulated; K_L is a named reduced-form estimand. Free parameters are pre-specified design knobs, not fitted elasticities. Load-bearing risk is concentrated in parallel trends given residual imbalance on the selection margin.

free parameters (4)
  • matched-overlap caliper on log pre-period fee revenue = 0.5 log-points (primary)
    Primary design uses 0.5 log-points (robustness at 0.25 and 1.0); determines which treated pools enter M and thus the estimand population.
  • event-study window = ±8 weeks (primary)
    Primary ±8-week window around pool-specific activation; robustness at ±6 and ±12. Horizon choice bounds the finite-horizon ATT path reported.
  • nearest-neighbor match multiplicity k = k≤3
    k≤3 nearest neighbors on pre-period realized fee revenue within caliper; affects control set and precision.
  • depth/liquidity band widths = ±1/2/5% (±2% primary)
    Active liquidity and depth measured in ±1%, ±2% (primary), ±5% bands around mid; outcome definition choice, fixed pre-estimation.
axioms (6)
  • domain assumption Dynamic conditional parallel trends in untreated increments for matched treated vs low-exposure controls after baseline-time adjustment (A3, Eq. 6).
    Required to read event-study coefficients β_h^m as ATT paths Δ_h^m; residual SMD 0.84 on fee revenue makes this the load-bearing identification assumption rather than level balance.
  • domain assumption Restricted interference: primary controls have near-zero exposure e_i≈0; exposed units are excluded from the primary counterfactual (A4).
    AMM pools share routers and pairs; without this, control outcomes could be contaminated by treated-pool spillovers.
  • domain assumption No clean-pre anticipation: proposal-to-activation weeks are excluded from the clean pre-period (A5).
    Governance proposal was public from November 2025; if anticipation already moved liquidity before activation, causal reading of post paths fails.
  • domain assumption Replayability: D_i, κ_i, t0_i, unit roles, and Y_it^m are deterministic functions of raw logs and fixed reconstruction rules (A1).
    On-chain design credibility requires that treatment and outcomes are not analyst-discretionary; paper asserts hash-checked frozen construction.
  • standard math Matched overlap: a treated unit enters M only if a low-exposure control exists within the pre-specified caliper on (S_i, X_i0) (A2).
    Standard common-support condition for matching estimators (Rosenbaum–Rubin / Stuart).
  • domain assumption Selection into the treatment list is on observables at the fee-revenue coverage margin, licensed by matching on pre-period realized fee revenue.
    Shock is protocol-wide and politics-timed, not randomized; quasi-experimental case rests on observable list-margin selection (Section 2.1).
invented entities (1)
  • K_L (LP-side liquidity-supply response kernel) independent evidence
    purpose: Shorthand for the finite-horizon reduced-form ATT path of active liquidity/depth (and related LP-supply outcomes) under the take-rate cut; the object fee-controller simulations often freeze.
    Not a new physical object; a named estimand (Eq. 2, M_A). Independent handle is the event-study path itself. Included because the paper elevates it as the design-based kernel separating LP-side from trader-facing claims.

pith-pipeline@v1.1.0-grok45 · 16949 in / 4025 out tokens · 43855 ms · 2026-07-10T05:58:11.756736+00:00 · methodology

0 comments
read the original abstract

Automated market maker (AMM) fee rules are often evaluated by liquidity-provider (LP) welfare, but that objective mixes fee revenue, adverse-selection loss (loss-versus-rebalancing, LVR), routing response, and liquidity supply. Fixed-fee Uniswap v3 history cannot separate these channels or identify counterfactual trader-facing dynamic-fee rules. Real fee-related variation nonetheless exists: the Uniswap protocol-fee switch cut LP take-rates with tier-differentiated intensity while leaving trader-facing fees unchanged. Using a pre-specified matched-overlap event-study difference-in-differences design, we estimate the liquidity-supply response to take-rate cuts, the kernel K_L that simulator-based fee-controller evaluations routinely freeze, while reconstructing treatment, event time, unit roles, and outcomes from public logs into a frozen, hash-checked panel before any estimate. We detect no large short-run average response in active liquidity or local depth; LP participation and composition, more precisely estimated, likewise show none, so the result is a non-detection at the design's resolution rather than a precise zero. Token-1 volume and native fee income fail the parallel-trends gate and are reported descriptively. A channel-admissibility audit delimits the estimand: the LP-side response K_L is design-based, while trader-facing dynamic-fee protection is a model-conditioned boundary, not a second estimand.

Figures

Figures reproduced from arXiv: 2607.08525 by Wen-Ting Wang.

Figure 1
Figure 1. Figure 1: Treatment activation and sample funnel. (a) Weekly [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Identification gate map for the outcome family. Tile colour is the causal disposition and the number [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Main event-study paths, primary ±8-week window, reference 𝑘 = −1, on the inverse-hyperbolic￾sine scale, with token-pair CR1 95% intervals. (a) Time-weighted active liquidity. (b) Local depth (±2% primary, with ±1% and ±5% as thin grey paths). Intervals include zero throughout the post period: a non-detection at the design’s resolution, not a precise zero. Disposition: identified null. 2020) the active-liqu… view at source ↗
Figure 4
Figure 4. Figure 4: Aggregate post-period ATT intervals (inverse-hyperbolic-sine scale) against the [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Matched-overlap balance and support. (a) Distribution of log pre-period realized fee revenue [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The channel map. Solid nodes are measured or reconstructed; the LP-side lane is activated [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

47 extracted references · 47 canonical work pages · 8 internal anchors

  1. [1]

    Review of Economic Studies , volume =

    Rambachan, Ashesh and Roth, Jonathan , title =. Review of Economic Studies , volume =

    work page

  2. [2]

    Journal of the Royal Statistical Society: Series B , volume =

    Cinelli, Carlos and Hazlett, Chad , title =. Journal of the Royal Statistical Society: Series B , volume =

    work page

  3. [3]

    and Samii, Cyrus , title =

    Aronow, Peter M. and Samii, Cyrus , title =. Annals of Applied Statistics , volume =

    work page

  4. [4]

    2021 , url =

    Adams, Hayden and Zinsmeister, Noah and Salem, Moody and Keefer, River and Robinson, Dan , title =. 2021 , url =

    work page 2021

  5. [5]

    2024 , url =

    Uniswap v4 Core , institution =. 2024 , url =

    work page 2024

  6. [6]

    am-AMM: An Auction-Managed Automated Market Maker

    Adams, Austin and Moallemi, Ciamac C. and Reynolds, Sara and Robinson, Dan , title =. arXiv preprint arXiv:2403.03367 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  7. [7]

    Fees in AMMs: A quantitative study

    Alexander, Abe and Fritz, Lars , title =. arXiv preprint arXiv:2406.12417 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  8. [8]

    Cryptoeconomic Systems , year =

    Angeris, Guillermo and Kao, Hsien-Tang and Chiang, Rei and Noyes, Charlie and Chitra, Tarun , title =. Cryptoeconomic Systems , year =

    work page

  9. [9]

    Optimal Dynamic Fees in Automated Market Makers , journal =

    Baggiani, Leonardo and Herdegen, Martin and S. Optimal Dynamic Fees in Automated Market Makers , journal =

    work page

  10. [10]

    Competition between DEXs through Dynamic Fees , journal =

    Baggiani, Leonardo and Herdegen, Martin and S. Competition between DEXs through Dynamic Fees , journal =

    work page

  11. [11]

    Optimal Fees for Liquidity Provision in Automated Market Makers

    Campbell, Steven and Bergault, Philippe and Milionis, Jason and Nutz, Marcel , title =. arXiv preprint arXiv:2508.08152 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  12. [12]

    Review of Financial Studies , year =

    Capponi, Agostino and Jia, Ruizhe , title =. Review of Financial Studies , year =

    work page

  13. [13]

    The Paradox Of Just-in-Time Liquidity in Decentralized Exchanges: More Providers Can Sometimes Mean Less Liquidity

    Capponi, Agostino and Jia, Ruizhe and Zhu, Brian Z. , title =. arXiv preprint arXiv:2311.18164 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  14. [14]

    Reinforcement Learning Approaches to Optimal Market Making , journal =

    Ga. Reinforcement Learning Approaches to Optimal Market Making , journal =

    work page

  15. [15]

    Optimal Dynamic Fees for Automated Market Makers: A Stochastic Control Approach to Loss-Versus-Rebalancing

    Ghasemlu, Farbod , title =. arXiv preprint arXiv:2606.21769 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  16. [16]

    Automated Market Making and Loss-Versus-Rebalancing

    Milionis, Jason and Moallemi, Ciamac C. and Roughgarden, Tim and Zhang, Anthony Lee , title =. arXiv preprint arXiv:2208.06046 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  17. [17]

    Pearl, Judea , title =

    work page

  18. [18]

    and Barto, Andrew G

    Sutton, Richard S. and Barto, Andrew G. , title =

    work page

  19. [19]

    von Stackelberg, Heinrich , title =

    work page

  20. [20]

    What Drives Liquidity on Decentralized Exchanges? Evidence from the Uniswap Protocol

    Zhu, Brian Z. and Liu, Dingyue and Wan, Xin and Liao, Gordon and Moallemi, Ciamac C. and Bachu, Brad , title =. arXiv preprint arXiv:2410.19107 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  21. [21]

    Encyclopedia of Epidemiology , editor =

    Greenland, Sander and Pearl, Judea , title =. Encyclopedia of Epidemiology , editor =

    work page

  22. [22]

    Proceedings of the Second Conference on Causal Learning and Reasoning (CLeaR) , publisher =

    Zhang, Chi and Mohan, Karthika and Pearl, Judea , title =. Proceedings of the Second Conference on Causal Learning and Reasoning (CLeaR) , publisher =

    work page

  23. [23]

    DEX Specs: A Mean Field Approach to DeFi Currency Exchanges

    Bayraktar, Erhan and Cohen, Asaf and Nellis, April , title =. arXiv preprint arXiv:2404.09090 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  24. [24]

    and Pischke, J

    Angrist, Joshua D. and Pischke, J. Mostly Harmless Econometrics: An Empiricist's Companion , publisher =

    work page

  25. [25]

    Estimation and Inference of Impulse Responses by Local Projections , journal =

    Jord. Estimation and Inference of Impulse Responses by Local Projections , journal =

    work page

  26. [26]

    Journal of the American Statistical Association , volume =

    Bojinov, Iavor and Shephard, Neil , title =. Journal of the American Statistical Association , volume =

    work page

  27. [27]

    and Rubin, Donald B

    Rosenbaum, Paul R. and Rubin, Donald B. , title =. Biometrika , volume =

    work page

  28. [28]

    , title =

    Stuart, Elizabeth A. , title =. Statistical Science , volume =

    work page

  29. [29]

    Roth, Jonathan and Sant'Anna, Pedro H. C. and Bilinski, Alyssa and Poe, John , title =. Journal of Econometrics , volume =

    work page

  30. [30]

    Journal of Econometrics , volume =

    Sun, Liyang and Abraham, Sarah , title =. Journal of Econometrics , volume =

    work page

  31. [31]

    American Economic Review: Insights , volume =

    Roth, Jonathan , title =. American Economic Review: Insights , volume =

    work page

  32. [32]

    , title =

    Liang, Kung-Yee and Zeger, Scott L. , title =. Biometrika , volume =

    work page

  33. [33]

    Colin and Miller, Douglas L

    Cameron, A. Colin and Miller, Douglas L. , title =. Journal of Human Resources , volume =

    work page

  34. [34]

    Colin and Gelbach, Jonah B

    Cameron, A. Colin and Gelbach, Jonah B. and Miller, Douglas L. , title =. Review of Economics and Statistics , volume =

    work page

  35. [35]

    Fast and Wild: Bootstrap Inference in Stata Using boottest , journal =

    Roodman, David and Nielsen, Morten. Fast and Wild: Bootstrap Inference in Stata Using boottest , journal =

    work page

  36. [36]

    Political Analysis , volume =

    Hainmueller, Jens , title =. Political Analysis , volume =

    work page

  37. [37]

    and Wichman, Casey J

    Bellemare, Marc F. and Wichman, Casey J. , title =. Oxford Bulletin of Economics and Statistics , volume =

    work page

  38. [38]

    and Magee, Lonnie and Robb, A

    Burbidge, John B. and Magee, Lonnie and Robb, A. Leslie , title =. Journal of the American Statistical Association , volume =

    work page

  39. [39]

    American Journal of Political Science , volume =

    Liu, Licheng and Wang, Ye and Xu, Yiqing , title =. American Journal of Political Science , volume =

    work page

  40. [40]

    2025 , url =

    R: A Language and Environment for Statistical Computing , author =. 2025 , url =

    work page 2025

  41. [41]

    Fast and User-Friendly Econometrics Estimations: The R Package fixest , journal =

    Berg. Fast and User-Friendly Econometrics Estimations: The R Package fixest , journal =. 2026 , doi =

    work page 2026

  42. [42]

    2021 , note =

    Fischer, Alexander and Roodman, David , title =. 2021 , note =

    work page 2021

  43. [43]

    2026 , note =

    Rambachan, Ashesh and Roth, Jonathan , title =. 2026 , note =

    work page 2026

  44. [44]

    2024 , note =

    Cinelli, Carlos and Ferwerda, Jeremy and Hazlett, Chad , title =. 2024 , note =

    work page 2024

  45. [45]

    2026 , note =

    Hainmueller, Jens , title =. 2026 , note =

    work page 2026

  46. [46]

    2026 , note =

    Liu, Licheng and Wang, Ye and Xu, Yiqing , title =. 2026 , note =

    work page 2026

  47. [47]

    fdapace: Functional Data Analysis and Empirical Dynamics , year =

    work page