samyaza-geek / YieldOptimizer

Streamlit app that backtests a DeFi yield optimization strategy. Uses a yield predictor and is mindful of slippage, gaz, yield dilution.

https://poolhopper.streamlit.app/

Streamlit app that backtests a DeFi yield optimization strategy. Uses a yield predictor and is mindful of slippage, gaz, yield dilution.

More details on https://medium.com/@nononymous/poolhopper-7fb6cf18182b

Architecture

1) Data collection (./scrappers)

a) ./scrappers/defillama_history (API to defillama)

First, filter DefiLama pools satisfying some criteria:

• a set of whitelisted protocols
• a set of whitelisted underlyings
• pool criteria (e.g. Tvl, chain, etc)

Then:

• collects data from Defillama.
• adjusts APY (eg Haircut rewards, add underlying APY (eg stETH), optionally assumes repegging and converts that into a yield).
• feeds a Postgres DB, either local or hosted by Aiden (name specified in params.yaml and auth in secrets.toml)

2) Feature engineering and modelling (./research)

• ResearchEngine class reads data from database, and provides methods to generate derived features (emwa, hvol, etc) as well as derived labels (multi horizon)
• provides methods to train and cross validate models.
• support most models in sklearn, as well as some custom models

3) bet sizing (./strategies/vault_rebalancing.py)

Performs portfolio optimization w/ transaction costs assuming a holding horizon. This performs a contrained convex optimization at each time step. All costs are annualized over am externally calibrated horizon.

4) Backtesting (./strategies/vault_backtest.py)

Backtests vault_rebalancing.py, with one or several sets of parameters

5) configs (./config)

• params.yaml is big yaml input parametrizing prediction, bet sizing, backtest etc..
• whitelist.yaml is the default list of protocols.
• grid.yaml is the default grid of parameters to backtest

6) streamlit UI (./main.py)

• leverages postgres DB to store data (./utils/postgres.py)
• authentification through telegram bot (./utils/telegram_bot.py)

guide

• to install, run pip install -r requirements.txt
• then run module streamlit run main.py to launch the streamlit app