Paper: arXiv 2509.05013
Authors: 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, \(L_t(x)\). 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.
Complexity vs Empirical Score
- Math Complexity: 8.5/10
- Empirical Rigor: 7.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced mathematical techniques like functional PCA, Legendre polynomials, and time-series models (AR(1)-GARCH) to analyze liquidity surfaces, indicating high math complexity. Empirical rigor is strong due to the use of real-world data from multiple Uniswap v3 pools, detailed statistical analysis, and robustness checks, though it lacks live trading implementation.
Research Flowchart
flowchart TD
A["Research Goal<br>Model Uniswap v3<br>Liquidity Dynamics"] --> B["Data Collection<br>3 Pools & Multiple Periods"]
B --> C["Methodology<br>Functional PCA<br>Dynamic Factor Models"]
C --> D{"Computational Analysis"}
D --> E["FPCA on L_t x<br>Extract Eigenfunctions"]
D --> F["Rolling Window Metrics<br>Time Series Factor Coeffs"]
E --> G["Key Findings & Outcomes"]
F --> G
G --> H["Stable Eigenfunctions<br>Legendre Polynomial Basis"]
G --> I["Factor Dynamics<br>AR(1) + GARCH<br>Heavy-tailed Inno."]
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper models the Uniswap v3 liquidity surface as a time‑tick surface using functional principal component analysis and dynamic factor models, aligns the empirical basis with low‑order Legendre polynomials, and proposes forecasting the factor coefficients with VAR‑GARCH methods.
Source sentence
This paper presents a comprehensive study on the empirical dynamics of Uniswap v3 liquidity, which we model as a time-tick surface, $L_t(x)$. Using a combination of functional principal component analysis (FPCA) and dynamic factor methods, we analyze three distinct pools over multiple sample periods.
Assumes 6
- For simplicity, assume the number of time observations T exceeds the number of tick points M and the covariance matrix is full rank. model
Source sentence
For simplicity, assume T > M and full rank, so that the eigenvalues can be ordered λ1 > λ2 > · · · > λM > 0.
- Assume the liquidity surface observations yt(x) are weakly stationary so that the population covariance has the same eigendecomposition as the sample covariance. statistical
Source sentence
If yt is weakly stationary with var(yt) = Σ = UΛU ⊤, then the same holds true at the population level: {βt,k}, UK = argmin {˜βt,k}, ˜UK E yt − K X k=1 ˜βt,k˜uk 2 .
- Assume the factor time series coefficients follow AR(1) dynamics with GARCH‑type heteroskedasticity and heavy‑tailed innovations. model
Source sentence
The factor coefficients exhibit a time series structure well-captured by AR(1) models with clear GARCH-type heteroskedasticity and heavy-tailed innovations.
- Assume block‑number spacing of approximately 8‑hour intervals (115200 blocks for ARB5 and 2400 for ETH) to approximate regular time spacing. data
Source sentence
We use a block-number spacing so that yt(x) and yt−1(x) are approximately 8 hours apart; this corresponds to 115200 blocks for ARB5 and 2400 for ETH.
Show 2 more
- Assume a large fraction of liquidity mass lies near the current price (x=0), implying the surface is peaked around the mid‑price. market
Source sentence
Seminal work in Uniswap v3 provides some preliminary stylized facts pertinent to our modelling approach [FMCM+21, FMCA+22]: • A large fraction of the liquidity mass lies near the current price, meaning we typically observe Lt(x) to be relatively large near x = 0.
- Assume the Legendre polynomial basis provides a fixed, interpretable set of factors comparable to level, slope, and curvature in yield‑curve models. model
Source sentence
This alignment provides a parsimonious and interpretable structure, similar to the dynamic Nelson-Siegel method for yield curves. This low-rank structure is highly stable across rolling time windows and also aligns consistently with a low-order Legendre polynomial basis.
Data
- Ethereum Mainnet 5 bps ETH‑USDC pool crypto-onchain proprietary
Source sentence
Our analysis incorporates data from a variety of Uniswap v3 pools to assess commonalities, generalizability, and improve robustness of takeaways: • Ethereum Mainnet 5bps ETH-USDC pool: This pool serves as our primary focus due to its consistently high trading volume and deep liquidity, Block spacing was chosen as 115200 for ARB5 and 2400 for ETH to obtain an (approximate) 8-hour time spacing.
- Ethereum Mainnet 30 bps ETH‑USDC pool crypto-onchain proprietary
Source sentence
Our analysis incorporates data from a variety of Uniswap v3 pools to assess commonalities, generalizability, and improve robustness of takeaways: • Ethereum Mainnet 5bps ETH-USDC pool: This pool serves as our primary focus due to its consistently high trading volume and deep liquidity, Block spacing was chosen as 115200 for ARB5 and 2400 for ETH to obtain an (approximate) 8-hour time spacing.
- Arbitrum 5 bps ARB‑USDC pool crypto-onchain proprietary
Source sentence
providing a rich dataset for in-depth analysis. • Ethereum Mainnet 30bps ETH-USDC pool: Included to investigate how dif-ferent fee tiers (and thus different ‘tickSpacing‘ values) impact liquidity distribution and dynamics, allowing for a comparison against the 5bps pool. • Arbitrum 5bps ARB-USDC pool: Included as a crucial robustness check to assess if observed features generalize across distinct blockchain
Costs
Not applicable
Finds 5
- The leading empirical eigenfunctions explain the majority of cross‑tick variation, remain stable over time, and align closely with a low‑order Legendre polynomial basis. statistical
Source sentence
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.
- Factor time‑series coefficients exhibit AR(1) dynamics with clear GARCH effects and heavy‑tailed innovations across all 5 bps pools. statistical
Source sentence
The factor coefficients exhibit a time series structure well-captured by AR(1) models with clear GARCH-type heteroskedasticity and heavy-tailed innovations. Our empirical analysis highlights a persistent low-rank factor representation within the Uniswap v3 liquidity surfaces of the 5 bps pools.
- The empirical PCA basis for the 5 bps pools is highly stable across rolling windows, supporting a parsimonious low‑rank representation. statistical
Source sentence
DYNAMICS OF LIQUIDITY SURFACES IN UNISWAP V3 29 A key finding is the stability of the empirical PCA basis for the 5 bps pools (Ethereum ETH-USDC and Arbitrum ARB-USDC). This low-rank structure is highly stable across rolling time windows and also aligns consistently with a low-order Legendre polynomial basis.
- The 30 bps ETH‑USDC pool shows a non‑stationary PCA basis that drifts substantially, requiring more components to capture the same variance. statistical
Source sentence
In contrast, the Ethereum 30 bps ETH-USDC pool revealed a highly nonstationary struc-ture with significant temporal irregularities. Its PCA basis drifted substantially over time, requiring a higher number of principal components to explain the same proportion of vari-ance observed in the 5 bps pools.
- A VAR‑GARCH forecasting recipe using the Legendre basis provides full distributional forecasts of the liquidity surface. predictive-accuracy
Source sentence
Consider the factor coefficient vector βt = [βt,1, . . . , βt,K]⊤. An appropriate model is βt+1 = a + Aβt + ϵt+1, ϵt+1|Ft ∼tν(0, Ht+1), (13) Here, a is an intercept vector, A contains the autoregression parameters, and Ht follows a multivariate GARCH specification.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 19 related papers). Coverage is partial, so this is neither replication nor refutation.
1 paper in the archive
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.