Dynamics of Liquidity Surfaces in Uniswap v3

Jimmy Risk, Shen-Ning Tung, Tai-Ho Wang · 2025

Read Dynamics of Liquidity Surfaces in Uniswap v3 by Jimmy Risk, Shen-Ning Tung, Tai-Ho Wang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-05013.txt.

Introduction

Liquidity on a decentralised exchange forms a surface that moves over time and price ticks. Jimmy Risk, lead author with two co-authors, study those dynamics for Uniswap v3 in a 2025 paper. They model liquidity as a time-tick surface and analyse three pools over several sample periods.

Functional principal component analysis and dynamic factor methods reduce the surface to a few interpretable components. For the tightest fee pools, the leading components explain most variation and stay stable, aligning with simple polynomial shapes. Their factor coefficients follow persistent time series patterns with heavy tails and changing volatility, giving a compact statistical picture of automated market maker liquidity.

Catalog record

CREATORJimmy Risk, Shen-Ning Tung, Tai-Ho Wang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE101 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-05013

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Dynamics of Liquidity Surfaces in Uniswap v3 Jimmy Risk, Shen-Ning Tung, Tai-Ho Wang Abstract . This paper presents a comprehensive study on the empirical dynamics of Uniswap v3 liquidity, which we model as a time-tick surface, L t ​ ( x ) L_{t}(x) . Using a combination of functional principal component analysis (FPCA) and dynamic factor methods, we analyze three distinct pools over multiple sample periods. Our findings offer three main contributions: a statistical characterization of automated market maker liquidity, an interpretable and portable basis for dimension reduction, and a robust analysis of liquidity dynamics using rolling window metrics. For the 5 bps pools, the leading empirical eigenfunctions explain the majority of cross-tick variation and remain stable, aligning closely with a low-order Legendre polynomial basis. This alignment provides a parsimonious and interpretable structure, similar to the dynamic Nelson-Siegel method for yield curves. The factor coefficients exhibit a time series structure well-captured by AR(1) models with clear GARCH-type heteroskedasticity and heavy-tailed innovations. 1. Introduction Decentralized finance (DeFi) [ HRS+21 , GM23 ] is rapidly transforming financial markets, with automated market makers (AMMs) playing a crucial role in this evolution. As a revolutionary force, decentralized exchanges (DEXs) [ CJ21 ] have emerged, attracting millions of users to DeFi projects and offering a way to bypass the constraints of traditional financial systems. Introduced by Uniswap Labs, Uniswap marked a pivotal shift within the DEX landscape by replacing traditional order-book mechanisms with its innovative AMM design. Version 1 enabled censorship-resistant spot trading, allowing swaps between Ether (ETH), Ethereum’s native asset, and any ERC-20 token (e.g., USDC, WBTC). Version 2 [ AZR20 ] generalized to any two ERC-20 tokens and added an on-chain oracle (a time-weighted average price). Version 3 [ AZS+21 ] introduced concentrated liquidity, allowing liquidity providers (LPs) to allocate capital only within a chosen price interval. Fees accrue from swaps executed when the current price lies within the LP’s defined range. Liquidity is a fundamental aspect of efficient markets, measuring the ease with which an asset can be exchanged without appreciable price impact [ BBDG18 ] . In centralized venues, liquidity is extensively studied through limit-order books (LOBs); queueing and stochastic partial differential equation (SPDE) models [ CST10 , HLR15 , CM21 ] effectively capture depth dynamics and resilience. By contrast, empirical work on AMM liquidity has so far centered on static range-allocation or fee-tier optimization. For example, Heimbach, Schertenleib, and Wattenhofer document that realized LP pay-offs hinge on tick-width and position lifetime [ HSW22 ] . Fan et al. [ FMCM+21 ] formalize the dynamic liquidity provision problem and focus on a general class of strategies to maximize LP earnings, and follow-up work provides a convex stochastic optimization problem that quantifies the optimal interval width for a belief-driven LP, providing implications for contract design [ FMCA+22 ] . Related is work on optimal adaptive allocation through deep reinforcement learning [ ZCY23 ] , equilibrium liquidity distribution via mean-field games [ BCN24 ] , and predictable-loss-aware optimal strategies derived from dynamic continuous time models [ CDM24 ] . On the more statistical side, panel regressions connect liquidity spreads and depth to gas costs, volatility, and fee revenue [ ZLW+24 ] ; researchers use linear models to forecast lookahead volume [ MC23 ] ; and statistical tests assess price discovery (across pools) and market efficiency [ ACDF25 ] . Complementary work measures impermanent loss and LP risk [ LHRW21 ] and clusters trader archetypes via graph embeddings [ MC24 ] . In summary, liquidity plays a crucial role in understanding larger aspects of DeFi. A better understanding of the liquidity surface L_{t}(x) aids comprehensive risk assessment, capital efficiency, and the development of sophisticated trading strategies. As mentioned, Uniswap v3 specifically generates a two-dimensional liquidity surface L_{t}(x) indexed by block-time t and tick distance x that represents the quantized price range within the Uniswap v3 architecture. In other words, each trading pair has its own temporal landscape, enabling empirical liquidity analysis, inference, and modeling. Surprisingly, we find that a thorough statistical description of the L_{t}(x) as a dynamic liquidity surface in (t,x) is still missing. This work is intended to fill that gap. In particular, we apply methods in dynamic factor models and functional principal component analysis (FPCA) to obtain a low-rank factor representation. This has been applied to find low-rank structures in other dynamic functional financial objects, such as yield curves [ Lit91 , CDR11 , Opr22 ] , implied volatility surfaces [ AHPP20 ] , and forward curves [ She09 , HSH12 ] . To motivate this approach, consider the well-known Nelson–Siegel (NS) factor model as introduced by Charles Nelson and Andrew Siegel (1987) [ NS87 ] . This was used as a parsimonious parametric framework to fit the term structure of interest rates (yield curves). The NS functional form uses three factors that are combined to approximate the cross-sectional shape of yields across maturities in a flexible but low-dimensional way [ DLY08 ] : (1) \hat{y}_{t}(\tau)=\beta_{1,t}\cdot 1+\beta_{2,t}\cdot\frac{1-e^{-\lambda\tau}}{\lambda\tau}+\beta_{3,t}\Big(\frac{1-e^{-\lambda\tau}}{\lambda\tau}-e^{-\lambda\tau}\Big), where where \hat{y}_{t}(\tau) is the model-implied yield at time t and maturity \tau , and \beta_{1,t} , \beta_{2,t} , \beta_{3,t} are time-varying coefficients (factors) while \lambda is a decay parameter that governs the shape. While the original model [ NS87 ] considered fixed coefficients (typically re-estimated), this dynamic factor model was a later extension by Diebold and Li (2006) [ DL06 ] to incorporate time-varying factors. This extends the cross-sectional specification to time-varying factors with their own dynamics. Despite its simplicity, this Dynamic Nelson–Siegel (DNS) model provides an excellent fit to the term structure at each point in time and often yields forecasts that outperform more theoretically complex models in out-of-sample interest rate forecasting. By projecting the entire yield curve onto three persistent factors and then modelling those factors with low-order autoregressions, Diebold and Li achieved a practical and accurate forecasting method for yields [ DL06 ] . Christensen et al. (2011) later connect with existing no-arbitrage models for yield curves to add a no-arbitrage link to DNS without sacrificing parsimony [ CDR11 ] . One of the remarkable aspects of the Nelson–Siegel basis is that its three latent factors naturally align with the principal components (PCs) observed in empirical yield curve data [ DL06 ] . Most of the variation in the yield curve can be explained by these three orthogonal factors, colloquially known as level, slope, and curvature, which persistently match the factors in the NS model. Specifically, a principal component analysis on a panel of interest rates across maturities reveals the first PC to be a level shift affecting all maturities, the second PC a slope factor that tilts the curve by raising short rates and lowering long rates or vice versa, and the third PC a curvature factor that bends the curve (hump-shaped, affecting medium maturities more than short or long ones). This aligns with findings by Litterman and Scheinkman (1991) [ Lit91 ] , who, through eigendecomposition of the covariance matrix of zero-return vectors, showed that the first factor (level) explains 89-90% of variance and is nearly constant across maturities; the second factor (slope/twist) explains about 8% and exhibits a monotone sign change; and the third factor (curvature/butterfly) explains 1-2% and affects medium maturities more. This low-rank structure is

Read the complete text