{"id":"7cb3921d-2633-42d7-b4a5-5a60688cf1f7","arxiv_id":"2501.07828","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Long positions in correlated risky pools with narrow ranges, and short wide-range positions in stable-risky pools, were the most profitable Uniswap v3 strategies in 2022 to 2024.","lead":"This paper analyzes 700 days of Uniswap v3 liquidity data and finds that LP profitability depends mainly on pool type, position duration, and position range size. It offers four simple strategies that on average beat the buy-and-hold baseline in some pools, but without gas fees and with survivorship caveats.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Strategy rankings may be an artifact of conditioning on realized duration: 'long-term' strategies are defined ex post, so the 2.69% edge is not an implementable ex-ante return.","rationale":"The reader correctly identified survivorship bias as a weakness, but the more load-bearing issue is that the strategy definitions themselves condition on realized duration, making the headline ranking not directly actionable. The paper's Section 5.3 acknowledges unclosed positions, yet does not address that even within closed positions, 'long-term' is an ex post label. This is a correctness risk for the practical contribution, not merely a data limitation. The paper is otherwise transparent, uses real Uniswap v3 data, and the IL measurement framework is standard, so the appropriate verdict remains CONDITIONAL rather than REJECT: the empirical findings could be salvaged by redefining strategies ex ante and re-estimating. I therefore recommend no change to the reader's conditional verdict, but with a sharper condition attached: the strategy-ranking claims must be re-verified under ex-ante definitions before being presented as LP guidance.","tokens_in":12057,"tokens_out":3376,"duration_ms":36856,"concrete_test":"Split the sample in half (e.g., May 2022-Apr 2023 for calibration, May 2023-Apr 2024 for evaluation). For every position opened in the evaluation window, fix range size at opening and assign an ex-ante holding rule (e.g., 'close after 30 days' or 'close after 1 day'); compute the strategy return on this rule using mark-to-market or actual close, rather than on realized duration. If long-term narrow-range no longer has the highest average return, or the ranking differs from Fig. 6, the headline claim is an artifact of conditioning on realized duration. As a secondary check, include all opened positions with unrealized IL at window end to quantify the acknowledged survivorship bias.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central ranking claim (Section 4.2, Fig. 6) is not implementable as stated because the four strategies are defined using realized position duration, which is only known after closure. Table 3 labels positions with duration >26.90 days as 'long-term' and <1.12 days as 'short-term'; these thresholds are estimated from the same sample (30th/70th percentiles) and then the same sample is used to measure returns. An LP cannot choose to be in the 'long-term narrow-range' bucket without knowing in advance that the position will survive 27 days, and the sample of closed positions with long durations is selected precisely because those positions were not closed after adverse price moves. This is a form of look-ahead/survivorship bias distinct from the unclosed-position bias acknowledged in Section 5.3: even among closed positions, short duration is often the result of stopping out after losses, while long duration selects for positions that avoided large IL early. Consequently the 2.69% average return for long-term narrow-range positions overstates the expected return of a strategy that commits to holding for 27 days, because such a commitment would include the paths that ended early. The in-sample percentile cutoffs compound this: thresholds derived from the full analysis window cannot be known when a strategy is chosen. The ranking could reverse under an ex-ante definition.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a measurement model based on loss-versus-holding (LVH) to compute realized impermanent loss and rewards for Uniswap v3 liquidity positions, applies it to 700 days of data from nine pools, and defines four liquidity provisioning strategies from position duration and range size. The central empirical claims are that the long-term narrow-range strategy yields the highest average returns (2.69%) in risky-risky pools and that the short-term wide-range strategy is the most profitable (0.14%) in stable-risky pools (Section 4.2). The paper concludes with parameter-level guidance for LPs on pool type, duration, and range size.","tokens_in":12360,"tokens_out":5410,"duration_ms":51235,"significance":"The paper addresses a practically important and timely question—how liquidity providers in concentrated-liquidity AMMs can improve profitability—and uses a substantial observational dataset. Its strengths include the use of a standard IL metric (LVH), a clear data collection and FIFO matching procedure, and an explicit acknowledgment of limitations such as survivorship bias and gas fees. If the strategy rankings were robust, they would offer actionable guidance. However, the central ranking is compromised by the ex-post definition of strategies based on realized position duration and in-sample percentile cutoffs, making the reported returns not directly implementable. The paper is a useful descriptive analysis, but the prescriptive conclusions require substantially more careful strategy definitions and robustness checks.","major_comments":[{"comment":"The four strategies are defined using realized position duration (e.g., long-term means duration > 26.90 days) and range-size cutoffs set at the 30th and 70th percentiles of the same sample on which performance is measured. Because duration is only known after a position is closed, the reported average return of 2.69% for the long-term narrow-range strategy is conditional on positions having survived for at least 27 days; it is not the expected return of an ex-ante strategy that commits to holding for 27 days. This is a look-ahead/survivorship bias distinct from the unclosed-position bias the authors acknowledge in Section 5.3. The in-sample percentile cutoffs compound the problem, since the thresholds are not knowable at the time a strategy is chosen. To support the prescriptive claim, the authors should redefine the strategies ex-ante (e.g., using intended holding periods or time-at-risk) or reframe Figure 6 as a descriptive, ex-post characterization rather than a ranking of implementable strategies.","section":"Section 4.2 / Table 3"},{"comment":"The paper excludes gas fees and acknowledges this in the limitations. This omission is load-bearing for the recommendation that short-term wide-range strategies are profitable in stable-risky pools, since the reported average return is only 0.14% with an average duration of 0.33 days (Section 4.2). For such short holding periods, network transaction costs and the costs of repeated position adjustments could easily exceed the gross return, possibly reversing the strategy ranking. The authors should quantify gas fees for representative position sizes or at least provide a break-even analysis so that LPs can judge whether the strategy is profitable after costs.","section":"Section 5.3"},{"comment":"Average returns are computed per position without weighting by position size or position value. Since the paper aims to guide LPs in allocating capital, value-weighted returns are more decision-relevant. The current unweighted averages can be dominated by many small positions, which may not reflect the experience of a typical capital-sized LP. A sensitivity analysis using value-weighted returns should be reported to establish whether the strategy rankings are robust to the weighting scheme.","section":"Section 4.2 / Figure 6"}],"minor_comments":[{"comment":"The FIFO matching procedure for deposit and withdrawal events is described only briefly. It would be helpful to state explicitly how rewards are allocated when the price leaves the position's range and how partial withdrawals are matched to earlier deposits.","section":"Section 3.3"},{"comment":"The text states that stable-risky pools exhibit higher IL than rewards and then reports negative average returns of -0.61%, while the figure shows distributions; please clarify whether these numbers refer to means or medians and how the statement relates to the figure.","section":"Section 4.1.1 / Figure 1"},{"comment":"Table 4 presents median daily IL and rewards, but the surrounding text appears to describe mean values. Please state clearly which statistic is being reported in each part of the table and text.","section":"Section 4.1.2 / Table 4"},{"comment":"The caption does not explain the subplot layout or the color/legend conventions. A brief description of the rows, columns, and what each panel displays would improve readability.","section":"Section 4.2 / Figure 6"},{"comment":"The threshold for excluding pool types with less than USD 10,000 in total value locked is not justified. Please provide a rationale or a sensitivity check showing that the findings are not sensitive to this threshold.","section":"Section 3.2"},{"comment":"The term 'adverse selection' is used to frame the measurement model, but the paper measures LVH, which compares a position to buy-and-hold rather than isolating adverse selection costs. A short clarification of the relationship between LVH and adverse selection would avoid conceptual confusion.","section":"Section 2.2"}],"recommendation":"major_revision","confidential_remarks":"The paper is within the scope of the journal and addresses a relevant topic. The main concern is the ex-post definition of strategies, which undermines the central prescriptive claim. The authors should either revise the strategy definitions or substantially temper their conclusions. I see no evidence of misconduct or citation issues; the reference list appears appropriate. I recommend major revision as described in the report."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a careful descriptive study of Uniswap v3 LP returns, and the data work is honest, but the headline claim about which strategy is most profitable is not implementable as stated. The four strategies are defined using realized position duration — 'long-term' means the position actually survived 27 days — so the 2.69% average return for long-term narrow-range positions is a description of positions that happened to last, not the return to a commitment to hold for 27 days. That is a real soft spot, and it is separate from the survivorship bias the authors acknowledge in Section 5.3.\n\nWhat is genuinely new: earlier work looked at duration or range size in isolation. This paper jointly classifies positions by both, across stable-risky and risky-risky pools, and reports returns for the four combinations. That is a useful empirical extension. The measurement model is standard LVH from prior work, but the execution is clean: nine pools, 700 days, FIFO matching of deposits and withdrawals, 95% confidence intervals, and a clear statement of limitations. Credit where due: they flag the unclosed-position survivorship bias, the exclusion of gas fees, and the focus on one DEX design.\n\nThe soft spots are mostly acknowledged, but the ex-post duration conditioning is not. Even among closed positions, short duration often means the position was stopped out after adverse price moves; long duration selects for positions that avoided early large IL. The percentile cutoffs (30th/70th of the same sample) compound the issue, since an LP cannot know these thresholds when choosing a strategy. Additionally, average returns are unweighted per position, so small positions count as much as large ones; gas fees would hit small positions hardest. These are not fatal for the paper's descriptive value — the finding that stable-risky pools have negative average returns, or that wide ranges reduce IL, is consistent with prior work and credible. But the practical guidance for LPs is weaker than the abstract implies.\n\nWho should read this: people working on AMM microstructure who want a recent, transparent dataset analysis of Uniswap v3 LP behavior. It is a solid descriptive contribution, not a breakthrough. I would send it to a serious referee — the data work deserves scrutiny and the strategy-ranking issue is fixable with an ex-ante definition or a clear reframing as descriptive correlation rather than actionable guidance.","headline":"Honest descriptive study of Uniswap v3 LP returns, but the headline strategy rankings are conditioned on realized duration and are not implementable ex ante.","tokens_in":12865,"tokens_out":1900,"would_cite":false,"duration_ms":18048,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Using 700 days of Uniswap v3 positions, this paper establishes that long-term narrow-range liquidity provisioning earns the highest average returns in risky-risky pools, while short-term wide-range provisioning is the most profitable in…","keywords":["automated market makers","liquidity provisioning strategies","impermanent loss","loss-versus-holding","Uniswap v3","decentralized finance","liquidity returns","adverse selection"],"falsifier":"Recompute average returns for the same pools after the analysis end date by also valuing still-open positions at current prices (unrealized IL plus accrued fees) and check whether the long-term narrow-range strategy still earns 2.69% and still beats the other three strategies.","tokens_in":11868,"feed_emoji":"📈","tokens_out":6179,"duration_ms":53680,"temperature":0.7,"pith_summary":"This paper tries to establish which choices liquidity providers can make in a concentrated-liquidity automated market maker (Uniswap v3) actually move their profits. Using 700 days of historical position data from nine pools, the authors build a measurement model based on loss-versus-holding to separate realized impermanent loss from fee rewards and then rank four parameter combinations. They find that long-term (over 26.9 days) narrow-range positions earn the highest average return of 2.69% in risky-risky pools, while in stable-risky pools the short-term (under 1.12 days) wide-range strategy is the most profitable at 0.14%. The broader claim is that pool type, position duration, range size, and position size each influence returns, with range size and duration being the levers LPs can practically adjust.","feed_headline":"Uniswap v3 data: long-term narrow-range wins in risky pools","feed_subtitle":"700 days of data: best strategy depends on pool correlation (2.69% risky-risky, 0.14% stable-risky).","key_machinery":"The carrying object is the loss-versus-holding measurement model: a formalization of impermanent loss for constant-product concentrated-liquidity AMMs, expressed by Equation (6), which gives realized IL as a function of the price change $d$ and the chosen lower and upper price bounds $p_a$ and $p_b$. The model compares the value of the deposited position with a buy-and-hold portfolio and computes LP returns as rewards minus realized IL. It is run on Uniswap v3 subgraph data, with positions matched deposit to withdrawal by FIFO, and strategies defined by cutting duration and range size at the 30th and 70th percentiles. This machinery lets the authors isolate each parameter's influence before comparing four strategy combinations.","core_discovery":"The central discovery is that liquidity provisioning returns on Uniswap v3 are systematically shaped by position duration and range size, and that the best strategy depends on pool token correlation. The authors' measurement model computes realized impermanent loss with the loss-versus-holding (LVH) metric and subtracts it from accumulated fee rewards to obtain LP returns. Applied to closed positions from May 2022 to April 2024, the model yields an average impermanent loss of -3.8% and shows that 49.5% of positions have negative returns. The strategy ranking is the paper's headline: long-term narrow-range positions average 2.69% returns in risky-risky pools, while short-term wide-range positions average 0.14% in stable-risky pools. From this the authors conclude that LPs should match duration and range size to the expected token price drift of the pool.","pith_inferences":["A testable extension is to correct for the survivorship bias the authors flag: value still-open positions at current prices with unrealized IL and accrued fees, then re-rank the four strategies; the long-term narrow-range advantage may shrink or disappear.","The finding that wide-range short-term positions beat narrow-range short-term positions in stable-risky pools suggests a repeatable tactical rule, but only if gas fees are negligible, since the paper's model excludes them.","The same measurement model could be applied to other AMM designs or to a later time window to test whether the parameter ranking generalizes beyond the May 2022 to April 2024 sample."],"forward_implications":["If correct, LPs in risky-risky pools should favor narrow ranges and long holding periods to capture concentrated-liquidity fee income while correlated token prices limit impermanent loss.","If correct, LPs in stable-risky pools should avoid narrow short-term positions and use wide ranges over short horizons to reduce adverse selection losses.","The result that fee rewards grow faster than IL with duration implies that closing positions early, particularly within the first day, locks in losses and is rarely profitable.","Pool type and fee tier act as the primary screens: stable-stable pools give low but positive returns, stable-risky pools are negative on average, and risky-risky pools are positive only with the right duration and range combination.","Strategy guidance can be made concrete: choose long-term narrow-range in risky-risky pools and short-term wide-range in stable-risky pools, and avoid treating position size as a profitability lever."],"supporting_citations":[{"why":"Defines the constant-product concentrated-liquidity mechanism and pool conventions used throughout the paper.","marker":"[1]"},{"why":"Supplies the loss-versus-holding metric that the measurement model uses to compute impermanent loss.","marker":"[2]"},{"why":"Provides the concentrated-liquidity impermanent loss formula adapted as Equation (6).","marker":"[19]"},{"why":"Provides the prior risk-and-return analysis of Uniswap v3 positions that the parameter analysis extends.","marker":"[20]"},{"why":"Motivates the pool-type and fee-tier analysis by showing how token composition shapes LP behavior and rewards.","marker":"[21]"},{"why":"Establishes the baseline that LPs often lose relative to holding, which the paper builds on and quantifies.","marker":"[28]"},{"why":"Describes loss-versus-rebalancing, the alternative metric that the paper sets aside in favor of LVH.","marker":"[31]"}],"fun_headline_variants":["Uniswap v3: Best LP strategy depends on token correlation","AMM LPs: Match range and duration to pool correlation","700 days of Uniswap v3 data: LP profits tied to correlation","Risky-risky pools: Long-term narrow-range yields 2.69%"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The ranking's load-bearing premise is that the closed positions in the sample fairly represent all LP outcomes; if unprofitable positions stay open and are excluded, the reported averages, especially the long-term returns, are too high.","fun_headline_variants_meta":{"raw":{"variants":["Uniswap v3: Best LP strategy depends on token correlation","AMM LPs: Match range and duration to pool correlation","700 days of Uniswap v3 data: LP profits tied to correlation","Risky-risky pools: Long-term narrow-range yields 2.69%"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001944,"raw_usage":{"total_tokens":7573,"prompt_tokens":883,"completion_tokens":6690,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":499,"completion_tokens_details":{"reasoning_tokens":6621}},"tokens_in":499,"tokens_out":6690,"duration_ms":45099,"temperature":1.0,"reasoning_tokens":6621,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T20:34:14.230593+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute average returns for the same pools after the analysis end date by also valuing still-open positions at current prices (unrealized IL plus accrued fees) and check whether the long-term narrow-range strategy still earns 2.69% and still beats the other three strategies.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the constant-product concentrated-liquidity mechanism and pool conventions used throughout the paper."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the loss-versus-holding metric that the measurement model uses to compute impermanent loss."},{"cited_title":"Moallemi, Tim Roughgarden, and Anthony Lee Zhang","cited_arxiv_id":null,"evidence_quote":"Describes loss-versus-rebalancing, the alternative metric that the paper sets aside in favor of LVH."}],"review_version":1}