Skip to content

Latest commit

 

History

History

lp-strategies

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
test
test
 
 
train
train
 
 
README.md
README.md
 
 
formula.png
formula.png
 
 

README.md

LP strategies

Last updated on 01/27/2022.

This dataset consists of the daily token price, reserve, and volume data of 34 different WETH pools in Uniswap V2, from 09/14/2020 to 01/26/2022. The list of tokens is as follows:

  • ALBT
  • CEL
  • COMP
  • DAI
  • DPI
  • FARM
  • HEX
  • KEEP
  • LEASH
  • LINK
  • LRC
  • MANA
  • MATIC
  • MKR
  • PERP
  • PICKLE
  • REN
  • RSR
  • SAND
  • SHIB
  • SNX
  • STAKE
  • STRONG
  • SUSHI
  • USDC
  • WBTC
  • AMP
  • AMPL
  • CRV
  • RENBTC
  • SWAP
  • USDT
  • XOR
  • YFI

The dataset has two versions, train and test. The train dataset spans from 09/14/2020 to 10/31/2021 whereas the test dataset contains the whole dataset, which means that the data from 11/01/2021 to 01/26/2022 can be only accessed within the test dataset. I did this just to avoid the look-ahead bias.

Period Frequency Start Date End Date Samples
Training Daily 09/14/2020 10/31/2021 413
Testing Daily 11/01/2021 01/26/2022 87

The observables are the daily token price, reserve, and volume for each token, which leads to a total of 3 x 16 = 48 columns. The targets are the estimated daily returns for each token pool, which were calculated by the following:

formula

where Fee Rate is given as 0.3 in Uniswap V2 and Price Change is defined as Price_t / Price_t-1. Due to the shift (or lag) operator involved in the calculation, the targets start from 09/14/2020 whereas the observables start from 09/13/2020 for the train dataset.

Benchmarks

Last updated on 01/29/2022.

Below shows the average performance for some benchmark models, aggregated from 10 different random seeds. The performance was measured as the sum of Root-Mean-Square Error (RMSE) from the targets of each token pool ignoring outliers. Outliers were defined as the data points that are further than 1.5 x IQR (Interquartile range) from the first and third quartiles. The performance was measured for the testing period (11/01/2021 - 01/26/2022).

Short-term Forecast - 1 Day

Model Name ΣRMSE Std
Previous Day 3.8291 -
Moving Average (10 days) 2.3202 -
Robust Regression + Stepwise OLS 1.7495 -
Moving Average (200 days) 1.7440 -
Prophet (default) 1.6217 -
Linear Regression (SGD) 1.4407 0.0024
Vector Autoregression (SGD) 1.4020 0.0008

Long-term Forecast - 10 Days

Model Name ΣRMSE Std
Previous Day 5.6349 -
Moving Average (10 days) 2.4263 -
Moving Average (200 days) 1.7545 -
Robust Regression + Stepwise OLS 1.7510 -
Prophet (default) 1.6684 -
Linear Regression (SGD) 1.4467 0.0019
Vector Autoregression (SGD) 1.4048 0.0005