📊 BTCUSDT October 2021 — TensorFlow Neural Network Strategy

Machine learning-based strategy that uses a TensorFlow neural network to predict next candle close prices.

The strategy trains a simple feed-forward neural network (8→6→4→1 architecture) on normalized candle data every 8 hours. It predicts where the next candle will close within its high-low range. When current price is below the predicted price, it opens a $100 position via Position.moonbag with 1% hard stop. Positions close automatically via trailing take profit when profit retraces by 1% from peak.

Strategy: oct_2021_strategy | Exchange: ccxt-exchange | Frame: oct_2021_frame

🚀 Quick Start

Link to the source code

npm start -- --backtest --symbol BTCUSDT ./content/oct_2021.strategy/oct_2021.strategy.ts
Copy

🧠 Neural Network Architecture

Input Layer:    8 neurons (8 normalized candles)
Hidden Layer 1: 6 neurons (ReLU activation, He Normal init)
Hidden Layer 2: 4 neurons (ReLU activation, He Normal init)
Output Layer:   1 neuron  (Sigmoid activation, outputs [0,1])
Copy

Training Parameters:

• Optimizer: Adam (learning rate 0.01)
• Loss: Mean Squared Error
• Epochs: 100
• Batch size: 32
• Validation split: 20%

Normalization: Each candle is normalized as (close - low) / (high - low), mapping the close position within the candle's range to [0, 1].

📉 Price Context (October 2021)

Metric	Value
Frame start	Oct 1, 2021
Frame end	Oct 14, 2021
Period	13 days
Timeframe	8h candles
Signal check	Every 15 minutes

✨ Performance Summary

Metric	Value
Total trades	28
Win trades	17
Loss trades	11
Win rate	60.71%
Total deployed capital	$2,800
Net PNL ($)	+$18.26
Net PNL (%)	+18.26%
ROI on capital	+0.65%
Avg PNL per trade	+$0.65 (+0.65%)
Best trade	+$5.37 (+5.37%)
Worst trade	−$1.40 (−1.40%)
Worst drawdown (%)	−1.40%
Worst drawdown ($)	−$1.40
Max consecutive wins	6
Max consecutive losses	4

📈 Risk Metrics

Metric	Value
Sharpe Ratio	0.312
Hard stop distance	1%
Trailing take distance	1% from peak
Max loss per position	$1 (1% of $100)
Avg trade duration	~8-24 hours

📋 Best & Worst Trades

Best Trade (+5.37%)

• Date: Oct 6, 2021 08:27 UTC
• Open: $50,901.21
• Close: $53,849.19
• Profit: +$5.37 (+5.37%)

Worst Trade (−1.40%)

• Date: Oct 13, 2021 06:00 UTC
• Open: $55,172.85
• Close: $54,621.12
• Loss: −$1.40 (−1.40%)

🔬 Strategy Logic

Signal Generation (every 15 minutes)

1. Training Phase (cached for 8h):

   • Fetch last 58 candles (50 for training + 8 for prediction window)
   • Train neural network on first 50 candles
   • Use last 8 candles as prediction input

2. Prediction:

   • Model outputs normalized close prediction [0, 1]
   • Convert to price: price = low + prediction * (high - low)

3. Entry Signal:

   if (currentPrice < prediction.price) {
     Position.moonbag({
       position: "long",
       currentPrice,
       percentStopLoss: 1.0,  // 1% hard stop
     })
   }
   Copy

4. Exit Logic (trailing take):

   • Monitors peak profit continuously
   • Closes position when profit retraces 1% from peak
   • Example: If position reaches +3% profit, closes at +2% (3% - 1% = 2%)

Position Management

• Entry size: $100 per trade
• Hard stop: 1% below entry
• Trailing take: 1% below highest profit peak
• Re-training: Model re-trains every 8h with fresh data

📊 Trade Distribution

Position Type	Count	Total PNL
LONG	28	+$18.26
SHORT	0	$0.00

🚀 How to Run

npm start -- --backtest --symbol BTCUSDT \
  --strategy oct_2021_strategy \
  --exchange ccxt-exchange \
  --frame oct_2021_frame \
  ./content/oct_2021.strategy/oct_2021.strategy.ts
Copy

Add --ui to open the web dashboard at http://localhost:60050:

npm start -- --backtest --symbol BTCUSDT --ui \
  ./content/oct_2021.strategy/oct_2021.strategy.ts
Copy

🌍 Environment Variables

Create a .env file in the project root (copy from .env.example):

# Telegram notifications (optional)
CC_TELEGRAM_TOKEN=your_bot_token_here
CC_TELEGRAM_CHANNEL=-100123456789

# Web UI server (optional, defaults shown)
CC_WWWROOT_HOST=0.0.0.0
CC_WWWROOT_PORT=60050
Copy