JonusNattapong/trading-gru-regression-xauusd

Trading GRU Regression Model for XAUUSD

This is a PyTorch GRU model trained to predict price change percentages for XAUUSD (Gold Futures).

Model Details

• Architecture: GRU with 3 layers, 128 hidden units, batch normalization, dropout
• Input: 50 timesteps of 16 technical indicators (standardized)
• Output: Predicted price change percentage (regression)
• Training Data: XAUUSD historical data from 2010-2023
• Loss: Mean Squared Error (MSE)
• Optimizer: Adam with L2 regularization
• Multi-Year Backtest Performance: 99.88% compounded return (19.98% average annual) across 2019-2024

Features Used

• Close, Volume, RSI_14, SMA_5, SMA_20, EMA_5, EMA_20
• MACD, MACD_Signal, MACD_Diff
• BB_Upper, BB_Lower, BB_Middle
• ATR_14, OBV, ROC_12

Usage

import torch
from sklearn.preprocessing import StandardScaler

class TradingLSTM(nn.Module):
    def __init__(self):
        super(TradingLSTM, self).__init__()
        self.gru = nn.GRU(input_size=16, hidden_size=128, num_layers=3, batch_first=True, dropout=0.3)
        self.fc1 = nn.Linear(128, 64)
        self.fc2 = nn.Linear(64, 32)
        self.fc3 = nn.Linear(32, 1)
        self.dropout = nn.Dropout(0.4)
        self.relu = nn.ReLU()
        self.batch_norm1 = nn.BatchNorm1d(128)
        self.batch_norm2 = nn.BatchNorm1d(64)

    def forward(self, x):
        gru_out, _ = self.gru(x)
        x = gru_out[:, -1, :]
        x = self.batch_norm1(x)
        x = self.relu(self.fc1(x))
        x = self.batch_norm2(x)
        x = self.dropout(x)
        x = self.relu(self.fc2(x))
        x = self.dropout(x)
        x = self.fc3(x)
        return x

model = TradingLSTM()
model.load_state_dict(torch.load('trading_regression.pth'))
model.eval()

# Prepare input sequence (50, 16) and scale with StandardScaler
# Predict price change percentage
prediction = model(sequence)  # e.g., 0.0167 = 1.67% expected change

Trading Strategy

• Buy when predicted change > 0.001 (0.1% expected increase)
• Sell when predicted change < -0.001 (0.1% expected decrease)
• Close positions when predictions reverse
• Tested across 5 years (2019-2024) with consistent profitability

Disclaimer

This model is for educational purposes only. Trading involves significant risk. Past performance does not guarantee future results.