This project analyzes how Market Sentiment (Fear & Greed Index) impacts trader behavior and profitability on the Hyperliquid exchange.
By merging 211k+ trade records with daily sentiment data, we uncover behavioral patterns, risk dynamics, and actionable trading rules.
The analysis focuses on:
- Performance differences across sentiment regimes
- Behavioral shifts in risk-taking and long/short bias
- Segmentation of traders into behavioral archetypes
- Actionable strategy design
Technologies: Python, Pandas, NumPy, Matplotlib, Seaborn, Scikit-Learn (K-Means)
Primetrade_Trader_Analysis.ipynb→ Main analysis notebookREADME.md→ Project overview and insights
- Open
Primetrade_Trader_Analysis.ipynbin Google Colab or Jupyter Notebook - Update dataset paths if needed
- Run all cells top-to-bottom
- Traders are predominantly buying (~51%) during Extreme Fear and selling during Extreme Greed (contrarian behavior).
- The highest average Closed PnL per trade (~$67.9) occurs during Extreme Greed, followed by Fear.
- Neutral markets are the least profitable.
- Losers: Bet their largest position sizes (~$8,190) during panic, often catching falling knives.
- Winners: Reduce size during downturns (
$4,414) and scale up aggressively ($9,612) only during Greed.
- The dataset does not contain an explicit leverage field.
- Position size (USD) is used as a proxy for risk exposure.
- Risk-taking increases during Greed and Extreme Greed.
Worst daily PnL per trader is used as a drawdown proxy, showing heavy negative tails for many traders, highlighting the importance of risk controls.
Using K-Means clustering on average PnL, win rate, position size, and trade count:
- 🐳 Whales – Very large sizes, moderate frequency
- 🎯 Snipers – Higher win rate, low frequency
- 🤖 Bots – Extremely high frequency, low PnL per trade
- 👥 Crowd – Low volume, small sizing, average performance
| Strategy | Trigger | Action | Logic |
|---|---|---|---|
| Anti-Panic Protocol | Index < 25 (Extreme Fear) | Hard cap position size to 50% | Prevents revenge trading and large losses |
| Trend-Surfer Protocol | Index > 60 (Greed) | Increase size limits + use trailing stop | Exploit most profitable regime |
- A Welch t-test between Fear and Greed PnL shows a directional difference but not statistically significant at 5% (p ≈ 0.064).
- Results should be interpreted as strong empirical patterns rather than guaranteed edges.
- Fear markets: higher positional buys and paradoxically high profits
- Greed markets: greatest profitability overall
- Losers increase risk during fear, winners during greed
For questions or collaboration, feel free to reach out.