Skip to content
This repository was archived by the owner on Apr 16, 2026. It is now read-only.
BranchesTags

Repository files navigation

📈 Stockbit Analysis - Whale Detection & AI Insights

A high-performance, containerized Go application for real-time stock market analysis, whale detection, and AI-powered pattern recognition using Stockbit data.

✨ Key Features

  • 🐋 Whale Detection: Real-time statistical anomaly detection (Z-Score > 3.0) to identify institutional activity with follow-up tracking.
  • 🧠 AI Insights: Integrated LLM agent (OpenAI-compatible) with intelligent pre-filtering and regime-adaptive confidence thresholds.
  • 📊 Signal History: Persistent quality tracking with regime-aware performance metrics.
  • ⚡ High Performance:
    • TimescaleDB: Efficient storage of millions of trade records with optimized candle aggregation.
    • Redis: Low-latency caching for baselines, regime data, and LLM results.
    • Go + SSE: Concurrent processing and real-time streaming to frontend.
  • 🔔 Notifications: Webhook integration for Discord/Slack alerts.

🎯 Enhanced Signal Generation

Multi-Layer Filtering Pipeline

Signal → RegimeFilter → StrategyPerformance → DynamicConfidence → OrderFlow → TimeOfDay → Position
           ↓                                                          ↓
      1.3x (trending)                                         1.5x (whale aligned)
      0.8x (ranging)                                          1.3x (strong buy)
      0.0x (volatile)                                         0.0x (whale divergence)

Regime-Based Optimization

Regime Characteristics LLM Threshold Position Multiplier
TRENDING_UP EMA slope > 0.5%, ATR < 2% 0.5 (relaxed) 1.3x (boost)
RANGING EMA slope < 0.5%, ATR < 2% 0.6 (default) 0.8x (reduce)
VOLATILE ATR > 2% 0.75 (strict) REJECT
TRENDING_DOWN EMA slope < -0.5% 0.7 (strict) 0.7x (reduce)

Whale Alignment Validation

Whale Activity Our Signal Action Multiplier
3+ BUY whales (>500M) BUY BOOST 1.5x
BUY > SELL BUY BOOST 1.3x
SELL > BUY (2+) BUY REJECT 0.0x

🧠 Logic at a Glance

Feature Threshold / Rule Action
Whale Detection Z-Score ≥ 3.0 AND Vol Spike ≥ 500% 🚨 ALERT
Regime Detection ATR-based with 5-min candles 📊 CLASSIFY
LLM Pre-filter Volume > 1000 lots, Value > 100M, Regime ≠ VOLATILE 🤖 ANALYZE
Volume Breakout Price > 2% AND Vol Z > 3.0 AND Trending 🟢 BUY
Whale Alignment 3+ BUY whales in 15min 🐋 BOOST 1.5x
Whale Divergence 2+ SELL whales vs BUY signal ⛔ REJECT
Stop Loss ATR-based (2× ATR) 🔴 CLOSE
Take Profit ATR-based (4× ATR TP1, 8× ATR TP2) 💰 CLOSE

🚀 Quick Start

  1. Setup Environment:

    cp .env.example .env
# Edit .env with your Stockbit credentials and LLM API key

  2. Configure Trading Parameters (Optional):

    # Regime-adaptive thresholds
TRADING_MIN_LLM_CONFIDENCE_TRENDING=0.5
TRADING_MIN_LLM_CONFIDENCE_VOLATILE=0.75

# LLM optimization
TRADING_MIN_VOLUME_FOR_LLM=1000
TRADING_MIN_VALUE_FOR_LLM=100000000
TRADING_LLM_COOLDOWN_MINUTES=3

  3. Run with Docker:

    make up

  4. Access Dashboard: Open http://localhost:8080

📊 Performance Metrics

Expected Improvements (vs baseline)

Metric Improvement
LLM API Costs -25-30%
Signal Win Rate +8-12% (50% → 58-62%)
Profit Factor +33-75% (1.2 → 1.6-2.1)
Max Drawdown -3% (-8% → -5%)
Sharpe Ratio +38-75% (0.8 → 1.1-1.4)

📚 Documentation

For detailed technical information, please refer to the docs/ directory:

🛠️ Project Structure

.
├── api/            # REST API & SSE Handlers
├── app/            # Core Application Logic
│   ├── regime_detector.go      # Market regime classification (ATR-based)
│   ├── signal_tracker.go       # Signal outcome tracking
│   ├── signal_tracker_gen.go   # LLM-based signal generation
│   ├── signal_filter.go        # Multi-layer signal filtering
│   ├── exit_strategy.go        # ATR-based exit levels
│   └── whale_followup_tracker.go
├── cache/          # Redis Caching Layer
├── config/         # Configuration Management
├── database/       # TimescaleDB Models & Repositories
├── docs/           # Documentation
├── llm/            # AI Agent Integration
├── public/         # Frontend Web UI
├── realtime/       # Real-time Broadcast System
└── ...

🔧 Advanced Features

Regime Detection

  • ATR Calculation: 14-period Wilder's smoothing on 5-minute candles
  • Trend Classification: EMA slope-based with 0.5% threshold
  • Volatility Measurement: ATR percentage (>2% = high volatility)
  • Confidence Scoring: Dynamic adjustment based on volatility

LLM Optimization

  • Pre-filtering: Skip volatile stocks, prioritize trending stocks
  • Dynamic Thresholds: 0.5 (trending) to 0.75 (volatile)
  • Caching: 5-minute TTL for analysis results
  • Cooldown: 3-minute per-symbol to prevent excessive calls

Signal Filtering

  • 5-Layer Pipeline: Regime → Strategy → Confidence → OrderFlow → TimeOfDay
  • Multiplier System: Combined multipliers up to 2.8x for perfect signals
  • Whale Validation: 15-minute window for institutional activity check
  • Auto-rejection: Volatile regime or whale divergence

🎓 Monitoring

Key Queries

Check regime distribution:

SELECT regime, COUNT(*), AVG(confidence)
FROM market_regimes
WHERE detected_at > NOW() - INTERVAL '1 hour'
GROUP BY regime;

Signal quality by regime:

SELECT mr.regime, 
       COUNT(so.id) as signals,
       ROUND(100.0 * SUM(CASE WHEN so.outcome_status = 'WIN' THEN 1 ELSE 0 END) / COUNT(so.id), 1) as win_rate
FROM signal_outcomes so
JOIN trading_signals ts ON so.signal_id = ts.id
LEFT JOIN market_regimes mr ON mr.stock_symbol = ts.stock_symbol
WHERE so.created_at > NOW() - INTERVAL '24 hours'
GROUP BY mr.regime;

🆕 Recent Updates

Enhanced Signal Quality (v2.0)

We've significantly improved signal quality through stricter filtering and better risk management:

Stricter Entry Criteria

Parameter Before After Impact
Require Order Flow false true Must have order flow confirmation
Buy Pressure Threshold 50% 55% Stronger buying confirmation
Aggressive Buy Threshold 55% 60% Higher smart money requirement
Min Baseline Samples 30 50 More historical data required
Low Win Rate Filter 40% 45% Faster rejection of underperforming strategies
Confidence Threshold 0.50 0.55 Higher signal quality

Improved Risk Management

  • Daily Loss Limit: Max 5% daily loss before trading stops
  • Circuit Breaker: Stops after 3 consecutive losses
  • Breakeven Protection: Triggers at 1% profit, moves stop to +0.15%
  • Fee-Aware Outcomes: Accounts for 0.25% round-trip fees

Time-Based Filters

  • Skip First 15 Minutes: Avoid 09:00-09:15 volatility
  • Pre-Lunch Caution: No signals 11:30-12:00
  • Post-Lunch Wait: Skip 13:30-13:45
  • Best Window: Priority for 10:00-11:00 signals

Read more: SIGNAL_IMPROVEMENTS.md

Swing Trading Support (NEW)

Hold positions overnight for larger profit potential!

Day Trading vs Swing Trading

Feature Day Trading Swing Trading
Holding Period Max 4 hours Max 30 days
Auto-Close 16:00 WIB ❌ No (hold overnight)
Stop Loss 1.5× ATR 4.5× Daily ATR
Take Profit 3×/6× ATR 9×/18× Daily ATR
Min Confidence 0.55 0.75
Min History 50 samples 400 samples (20 days)
Trend Required Above VWAP Strong trend (score > 0.6)

Swing Trade Criteria

A signal qualifies as swing trade if:

  1. ✅ Confidence ≥ 0.75
  2. ✅ 20+ days of historical data
  3. ✅ Trend score ≥ 0.6
  4. ✅ Swing Score ≥ 0.65 
    Swing Score = (Confidence × 0.4) + (Trend × 0.4) + (Volume × 0.2)

Configuration

# Enable Swing Trading
SWING_TRADING_ENABLED=true
SWING_MIN_CONFIDENCE=0.75
SWING_MAX_HOLDING_DAYS=30
SWING_ATR_MULTIPLIER=3.0
SWING_MIN_BASELINE_DAYS=20
SWING_POSITION_SIZE_PCT=5.0
SWING_REQUIRE_TREND=true

Read more: SWING_TRADING.md

📊 API Reference

Signal Statistics Endpoint

Debug signal flow and filtering:

GET /api/signals/stats?lookback=60

Response:

{
  "total_signals": 50,
  "by_decision": {"BUY": 5, "WAIT": 20, "NO_TRADE": 25},
  "by_outcome_status": {"OPEN": 2, "SKIPPED": 45, "PENDING": 3},
  "truly_pending": 3
}

Position Endpoints

  • GET /api/positions/open - View open positions
  • GET /api/positions/history - View closed positions with P&L
  • GET /api/signals/history - Full signal history

Analytics Endpoints

  • GET /api/analytics/strategy-effectiveness - Performance by strategy
  • GET /api/analytics/optimal-thresholds - Best confidence levels
  • GET /api/analytics/time-effectiveness - Best trading hours
  • GET /api/analytics/expected-values - EV calculations

License

This project is for educational purposes only. Not for financial advice.

About

No description, website, or topics provided.

Resources

Readme
Activity

Stars

29 stars

Watchers

0 watching

Forks

10 forks
Report repository

Releases

Packages

Contributors

Languages