charieswei
/
TradingAgents
mirror of https://github.com/TauricResearch/TradingAgents.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
TradingAgents/PUMP_DETECTION_GUIDE.md

9.7 KiB
Raw Blame History

Pump Detection System

Overview

The Pump Detection System identifies stocks likely to experience sudden price increases (pumps) BEFORE they happen. This gives you early entry opportunities to profit from momentum moves.

What is a "Pump"?

A pump is a rapid, significant increase in stock price driven by:

  • Volume surge - Abnormal trading activity (2-10x average volume)
  • Price acceleration - Quick gains over 1-3 days (5-50% moves)
  • Social buzz - Heavy discussion on Reddit, Twitter, StockTwits
  • Technical setup - Stocks bouncing from oversold conditions (RSI < 30)
  • Catalysts - Upcoming earnings, FDA approval, partnership news

The key to profiting is identifying these signals BEFORE the pump starts, not chasing after it's already up 50%.

How the Detection Works

Detection Methods

The system uses 5 complementary detection strategies:

1. Volume Spike Detection

  • What it detects: Abnormal increase in trading volume
  • Signal: Volume > 2x average (configurable)
  • Interpretation: Increased institutional/retail interest, often precedes price moves
  • Tool: detect_volume_spike(symbol, curr_date, look_back_days=20, threshold_multiplier=2.0)

2. Price Acceleration

  • What it detects: Rapid price gains in recent period
  • Signal: Recent gains > 5% AND recent gains > 1.5x older gains
  • Interpretation: Momentum building, potential pump in progress or starting
  • Tool: detect_price_acceleration(symbol, curr_date, look_back_days=10)

3. Social Sentiment Surge

  • What it detects: Spike in social media mentions and sentiment
  • Signal: Stock in top trending list on Reddit/Twitter/StockTwits
  • Interpretation: Retail hype building, retail money may follow
  • Tool: detect_social_sentiment_surge(symbol, curr_date)

4. Oversold Bounce Setup

  • What it detects: Stocks bouncing from oversold technical conditions
  • Signal: RSI < 30 (indicator of oversold)
  • Interpretation: High probability bounce/reversal, often with strong moves
  • Tool: detect_oversold_bounce(symbol, curr_date, rsi_threshold=30)

5. Catalyst Events

  • What it detects: Upcoming earnings, FDA approvals, partnerships, etc.
  • Signal: Major event within 3 months
  • Interpretation: Catalyst drives speculation and trading volume
  • Tool: detect_catalyst_event(symbol, curr_date)

Pump Score Calculation

All signals are combined into a Pump Probability Score (0-100):

Score = 25% (Volume) + 20% (Price) + 15% (Social) + 20% (Technical) + 20% (Catalyst)

Score Interpretation:

  • 70+: 🔴 VERY HIGH pump probability → Strong entry candidate
  • 50-69: 🟠 HIGH pump probability → Good entry with risk management
  • 30-49: 🟡 MODERATE pump probability → Wait for confirmation
  • <30: 🟢 LOW pump probability → No pump signals detected

Usage

Quick Demo (with cached data)

python pump_detection_demo.py

This analyzes cached stock files and shows pump detection scores.

Live Pump Screening (requires data sources)

Analyze specific stock:

python pump_screening.py --ticker NVDA

Analyze specific date:

python pump_screening.py --ticker TSLA --date 2025-12-05

Full market screening (find pump candidates):

python pump_screening.py

In Your Trading Agent Code

from tradingagents.agents.pump_detection_agent import create_pump_detection_agent
from tradingagents.agents.utils.pump_detection_tools import (
    detect_volume_spike,
    detect_price_acceleration,
    detect_social_sentiment_surge,
    detect_oversold_bounce,
    detect_catalyst_event,
    calculate_pump_score,
)

# Create agent
pump_detector = create_pump_detection_agent(llm)

# Use in workflow
state = {
    "messages": [],
    "ticker": "NVDA",
    "trade_date": "2025-12-05",
}
result = pump_detector(state)

Pump Trading Strategy

Entry Rules

  1. Pump Score >= 50: Consider entry
  2. Multiple signals confirmed: Volume + Price Acceleration = strong signal
  3. Early position: Enter in first 5-15% of move
  4. Small size: 1-2% of portfolio max per trade

Position Management

Entry Price: $100
Stop Loss: $97-98 (2-3% below entry)
Target 1: $105 (5% profit)
Target 2: $110 (10% profit)
Target 3: $115+ (15%+ profit)

Exit Rules

  • ❌ Exit on Stop Loss: Hard stop at 2-3% loss
  • ❌ Exit on Volume Collapse: If volume drops 50%, pump ending
  • ❌ Exit at Resistance: Price hitting resistance level
  • ✅ Exit at Targets: Take profits at 5-10% gains
  • ⏸️ Trail Stop: Move stop to breakeven once +5% profit achieved

Risk Management

⚠️ Critical Rules

  1. Position Size: Never more than 1-2% per pump trade
  2. Stop Loss: ALWAYS set at entry, no exceptions
  3. Risk/Reward Ratio: Min 1:2 (risk $1 to make $2)
  4. Diversification: Max 3-4 pump trades simultaneously
  5. Account Allocation: Max 5-10% of portfolio in pump trades total

Avoid These

  • 🚫 Chasing after 50% gains - Too late, dumpers coming
  • 🚫 No stop loss - Pump can reverse quickly
  • 🚫 All-in positions - One bad pump wipes account
  • 🚫 Holding overnight - Pumps can reverse in minutes
  • 🚫 Ignoring volume - Volume collapse = pump ending
  • 🚫 Fighting the trend - Don't short pumps, ride them

Pump vs Pump-and-Dump Warning Signs

Legitimate Pump (Good for trading)

  • ✅ Supported by volume surge
  • ✅ Following technical setup (oversold bounce)
  • ✅ Has catalyst event
  • ✅ Social sentiment building gradually
  • ✅ Price holds gains or consolidates

Pump-and-Dump (Dangerous - AVOID)

  • ⚠️ Extreme volume spike (5-10x+)
  • ⚠️ Penny stock with low liquidity
  • ⚠️ Coordinated social media push
  • ⚠️ Insider selling after spike
  • ⚠️ No fundamental reason for move
  • ⚠️ Stock drops 50%+ within days

Protection: Use the detect_pump_and_dump_risk() tool before entry.

Available Tools

Core Detection Tools

Tool Purpose Usage
detect_volume_spike() Find abnormal volume Identify interest surge
detect_price_acceleration() Find rapid gains Confirm momentum building
detect_social_sentiment_surge() Find social buzz Track retail interest
detect_oversold_bounce() Find technical setup Identify bounce candidates
detect_catalyst_event() Find catalysts Find event-driven moves
calculate_pump_score() Calculate probability Score all signals

Agent

Agent Purpose
create_pump_detection_agent() Full pump detection analysis
create_screening_agent() Market screening for pump candidates

Examples

Example 1: High Probability Pump

🎯 Stock: NVDA

✅ Volume Spike: 3.2x average - DETECTED
✅ Price Acceleration: +12% in 2 days - DETECTED
✅ Social Buzz: Top 5 trending on Reddit - DETECTED
❌ Oversold Bounce: RSI 65 - Not oversold
✅ Catalyst: Earnings next week - DETECTED

Pump Score: 78/100 (VERY HIGH)
Recommendation: STRONG BUY SIGNAL
Entry: Now with 2% stop loss
Targets: $1000, $1050, $1100+

Example 2: Weak Pump Signal

🎯 Stock: XYZ

❌ Volume Spike: 1.1x average - Not detected
❌ Price Acceleration: +0.5% - Not detected
✅ Social Buzz: Mentioned on social - DETECTED
✅ Oversold Bounce: RSI 25 - DETECTED
❌ Catalyst: None upcoming - None

Pump Score: 35/100 (MODERATE - WAIT)
Recommendation: HOLD - Wait for volume confirmation
Monitor: Watch for volume spike as entry trigger

Troubleshooting

"All vendor attempts failed"

  • Ensure API keys are set (Alpha Vantage)
  • Or ensure cached data exists
  • Run python main_screening.py first to cache data

"Insufficient data"

  • Stock needs at least 5 days of historical data
  • Use longer look_back_days for new stocks

No social buzz detected

  • API rate limits reached
  • Stock may not be discussed enough
  • Try again after cool-down period

Best Practices

  1. Always backtest your pump detection settings
  2. Start small with 0.5-1% position size
  3. Monitor continuously - don't leave pump trades unattended
  4. Keep a trading journal - track which signals work best
  5. Use trailing stops - protect profits as pump develops
  6. Combine with other analysis - don't rely solely on pump detection
  7. Review losses - learn from failed pump trades

Advanced Customization

Adjust Detection Thresholds

from tradingagents.agents.utils.pump_detection_tools import detect_volume_spike

# More aggressive (2x volume threshold)
result = detect_volume_spike.invoke({
    "symbol": "NVDA",
    "curr_date": "2025-12-05",
    "threshold_multiplier": 2.0,  # Default
})

# More conservative (3x volume threshold)
result = detect_volume_spike.invoke({
    "symbol": "NVDA",
    "curr_date": "2025-12-05",
    "threshold_multiplier": 3.0,
})

Custom Score Weights

Edit tradingagents/agents/utils/pump_detection_tools.py:

weights = {
    "volume_spike": 40,  # Increase from 25 to 40
    "price_acceleration": 20,
    "social_sentiment": 10,  # Decrease from 15 to 10
    "oversold_bounce": 20,
    "catalyst": 10,  # Decrease from 20 to 10
}

Support & Questions

For issues or questions:

  1. Check the troubleshooting section above
  2. Review the examples
  3. Check cached data files exist in tradingagents/dataflows/data_cache/
  4. Review error messages in terminal output

Disclaimer

⚠️ Pump trading is HIGH RISK. Stocks can:

  • Drop as fast as they rose
  • Get halted by SEC
  • Be delisted
  • Never recover

This system is for experienced traders only. Use strict risk management and never trade more than you can afford to lose.

Past performance does not guarantee future results.