Abstract:
Algorithmic trading has revolutionized financial markets for amount of time by leveraging machine learning and statistical models to optimize investment strategies, especially in cryptocurrency which explicit the most volatile and fluctuate market nowadays. This study investigates the performance of hybrid algorithmic trading strategies in cryptocurrency markets by integrating Long Short-Term Memory (LSTM) networks, XGBoost models, and portfolio management techniques compare with standalone algorithm. The research utilizes historical market data, volume, market capitalization and number of transactions from Bitcoin (BTC), Ethereum (ETH), Ripple coin (XRP), Binance Coin (BNB), and Litecoin (LTC) to develop predictive algorithm models for price movement. The study employs feature engineering techniques to preprocess financial indicators such as Moving Averages, Relative Strength (RS), Relative Strength Index (RSI), On balance volume (OBV), Awesome oscillator (AO), Bollinger Bands, and Fibonacci while incorporating machine learning models for price forecasting. Additionally, incorporating portfolio management strategies, including the Kelly Criterion, Sharpe Ratio, optimized drawdown, stop loss, and trailing stop, are integrated into the trading framework to enhance profitability while minimizing potential losses. The models performance is evaluated based on key metrics, including accuracy, recall, root mean squared error (RMSE), and return on investment (ROI). Results of the study indicate that hybrid algorithmic trading strategies outperform traditional standalone models, demonstrating improved predictive accuracy and profitability in volatile cryptocurrency markets. Additionally, hybrid model with portfolio management integrated show better result than hybrid model without other techniques included. The study highlights the potential of integrating machine learning with financial portfolio management to enhance decision-making in algorithmic trading.