Bitcoin Market Sentiment × Trader Performance Analysis

A quantitative study correlating the Crypto Fear & Greed Index with live trader behavior on the Hyperliquid exchange — and a data-driven strategy that improved per-trade profitability by 57.4%.

---

Table of Contents

• Overview
• Key Results
• Dataset
• Strategy Logic
• Analysis Highlights
• Repository Structure
• Getting Started
• Visualizations
• Methodology
• Future Work

---

Overview

This project explores a fundamental question in crypto trading: does market sentiment measurably affect trader profitability, and can we exploit that relationship?

By merging 211,000+ trade records from the Hyperliquid perpetuals exchange with daily Fear & Greed Index readings, this analysis uncovers clear behavioral patterns across sentiment regimes — and engineers a filtering strategy that systematically avoids low-signal environments.

---

Key Results

Metric	Value
Total Trades Analyzed	211,000+
Baseline Avg. Closed PnL	$48.55
Strategy Avg. Closed PnL	$76.43
Performance Lift	+57.4%
Worst Sentiment Regime	Neutral ($34.31 avg PnL)

Bottom line: Traders who avoid "Neutral" sentiment windows and align direction with extreme sentiment phases capture significantly higher per-trade returns.

---

Dataset

Source	Description
historical_data.csv	Raw trade records from Hyperliquid — includes timestamp, direction, size, and Closed PnL
fear_greed_index.csv	Daily Fear & Greed Index readings (0–100) with categorical labels
submission_results.csv	Merged and enriched output with sentiment classifications per trade

Sentiment categories used: Extreme Fear · Fear · Neutral · Greed · Extreme Greed

---

Strategy Logic

The Smart Sentiment Engine applies three counter-cyclical rules derived from the data:

1. BUY  →  Prioritized during [ Extreme Fear ] and [ Fear ]
              Contrarian accumulation when the crowd panic-sells.

2. SELL →  Prioritized during [ Extreme Greed ]
              Profit realization ahead of potential mean reversion.

3. SKIP →  All activity filtered out during [ Neutral ]
              Noise reduction — Neutral regimes show the lowest
              signal-to-noise ratio across all sentiment bands.

This is a signal filter, not a standalone trading system. It is designed to be layered on top of any existing directional strategy to improve regime-adjusted performance.

---

Analysis Highlights

Sentiment vs. Average PnL

Sentiment	Avg Closed PnL	Signal Quality
Extreme Fear	High	✅ Strong
Fear	Above Average	✅ Good
Neutral	$34.31	❌ Weak
Greed	Above Average	✅ Good
Extreme Greed	High	✅ Strong (SELL bias)

The "Neutral Trap"

Neutral sentiment is not a safe middle ground — it produces the lowest average PnL of any category. This is attributed to elevated market noise and indecisive price action, resulting in a higher rate of marginal or losing trades. The strategy's single highest-impact rule is simply: stop trading when nobody has conviction.

---

Repository Structure

.
├── solutions.py              # Core processing script and strategy simulation
├── submission_results.csv    # Merged dataset with sentiment classifications
├── pnl_sentiment.png         # PnL distribution across sentiment categories
├── historical_data.csv       # (Required) Raw Hyperliquid trade records
├── fear_greed_index.csv      # (Required) Daily Fear & Greed Index data
└── README.md

---

Getting Started

Prerequisites

• Python 3.8+
• pandas, matplotlib, seaborn

Installation

pip install pandas matplotlib seaborn

Data Setup

Place the following files in the project root before running:

historical_data.csv
fear_greed_index.csv

Run

python solutions.py

The script will:

1. Load and merge both data sources on normalized timestamps
2. Classify each trade by its corresponding sentiment regime
3. Compute baseline vs. strategy-filtered PnL statistics
4. Output submission_results.csv and pnl_sentiment.png

---

Visualizations

pnl_sentiment.png — Distribution of average Closed PnL across each sentiment category, highlighting the performance gap between high-signal (Extreme Fear / Extreme Greed) and low-signal (Neutral) regimes.

---

Methodology

Step	Detail
Data Merging	Timestamp normalization to daily granularity via pandas; left-join on trade date
Sentiment Binning	Fear & Greed Index mapped to 5 standard categories per Alternative.me classification
Baseline	Mean Closed PnL across all 211K+ unfiltered trades
Strategy Filter	Subset trades matching directional + sentiment criteria; exclude Neutral regime entirely
Evaluation Metric	Mean Closed PnL per trade (strategy vs. baseline)

---

Future Work

• Intraday sentiment granularity — Hourly Fear & Greed proxies (funding rates, CVD) for sub-daily filtering
• Per-asset breakdown — Sentiment effectiveness may vary across BTC, ETH, and altcoin pairs
• Position sizing overlay — Scale size proportionally to sentiment extremity, not just binary on/off
• Backtested equity curve — Convert per-trade PnL lift into a time-series P&L curve with drawdown analysis
• Live signal integration — API hook to Alternative.me for real-time regime classification

---

Disclaimer

This project is for research and educational purposes only. Past performance of any trading strategy does not guarantee future results. Nothing in this repository constitutes financial advice.