Trader Performance vs Market Sentiment

Objective

Analyze how Bitcoin market sentiment, measured by the Fear/Greed Index, relates to Hyperliquid trader behavior and performance.

Project Structure

.
├── analyze_trader_sentiment.py
├── data/
│   ├── fear_greed_index.csv
│   └── historical_data.csv
├── charts/
│   ├── avg_daily_pnl_by_sentiment.svg
│   ├── trade_frequency_by_sentiment.svg
│   ├── segment_pnl_per_trade.svg
│   └── daily_total_pnl.svg
├── outputs/
│   ├── data_profile.csv
│   ├── sentiment_performance_summary.csv
│   ├── segment_summary.csv
│   ├── daily_account_metrics.csv
│   ├── daily_market_metrics.csv
│   ├── account_segments.csv
│   └── top_accounts_by_pnl.csv
└── REPORT.md

How to Run

1. Install dependencies:

pip install pandas numpy

2. Place the two source CSV files in data/:

data/fear_greed_index.csv
data/historical_data.csv

3. Run the analysis:

python analyze_trader_sentiment.py

The script regenerates all tables in outputs/, charts in charts/, and the written summary in REPORT.md.

Methodology

• Standardized raw column names and parsed trader timestamps from Timestamp IST.
• Created a daily date key and joined trades to the Bitcoin Fear/Greed dataset by date.
• Aggregated trade-level data to account-day level to reduce bias from very active accounts.
• Created performance metrics:
  • daily PnL
  • win rate
  • PnL per trade
  • PnL per $1k traded volume
  • downside-PnL drawdown proxy
• Created behavior metrics:
  • trades per account-day
  • average trade size
  • long/short share
  • average position exposure proxy
• The source trader dataset does not contain an explicit leverage column, so abs(start_position) * execution_price is used as an exposure proxy.

Key Findings

1. Average daily account PnL was highest on Fear days and weakest on Greed days.
2. Fear-side days had materially higher trade frequency than Greed-side days.
3. Greed-side days had a higher average win rate, but lower average PnL than Fear-side days.
4. The raw sentiment score has weak direct correlation with daily PnL, so it is better used as a risk-regime feature than as a standalone signal.
5. Segmenting traders improves the story: infrequent traders had better PnL per trade than frequent traders, and consistent winners had better PnL per trade than inconsistent traders.

Strategy Recommendations

1. Use sentiment as a risk filter. In weaker-performing regimes, especially Greed in this sample, reduce exposure and require stronger confirmation before increasing trade frequency.
2. Apply segment-specific allocation. Prioritize trader archetypes with better PnL per trade, especially infrequent traders and consistent winners, while monitoring whether the advantage persists across sentiment regimes.

Main Outputs

• REPORT.md: concise written summary for submission.
• outputs/sentiment_performance_summary.csv: performance and behavior by sentiment classification.
• outputs/segment_summary.csv: account segment comparison.
• charts/avg_daily_pnl_by_sentiment.svg: core performance chart.
• charts/trade_frequency_by_sentiment.svg: behavior shift chart.
• charts/segment_pnl_per_trade.svg: segment comparison chart.