Resumo

Cryptocurrencies have emerged as a highly volatile and fast-growing asset class, attracting investors, speculators, and researchers (Sebastião & Godinho, 2021). The volatility and speculative nature of cryptocurrencies such as Bitcoin, Ethereum, and others present significant challenges and unique opportunities for applying advanced price prediction models and trading strategies. Existing literature often focuses on predictive models' error metrics but lacks practical applications in trading and portfolio management.

This study addresses gaps in the literature by proposing an integrated approach that connects classification metrics of price predictions with investor returns, conducts investment simulations, and explores diversification techniques for risk minimization. The main objective is to demonstrate the relationship between forecast accuracy and actual returns, evaluating how different classification metrics affect trading strategy performance through backtesting and risk management techniques.

The literature highlights the effectiveness of machine learning models, including stacking techniques, in predicting cryptocurrency returns. Stacking integrates multiple information sources and predictive techniques, resulting in robust trading strategies that adapt to the volatile dynamics of the cryptocurrency market. Studies have shown that models combining technical and macroeconomic data achieve higher accuracy and better performance in financial predictions.

The methodology involves predicting cryptocurrency price movements using a stacking approach of machine learning models. Data from the Binance exchange included technical indicators and market variables for various cryptocurrencies. The stacked models aim to capture different market patterns and dynamics, utilizing cross-validation for training and testing. The study employed base models such as XGBoost, Random Forest, LightGBM, and Support Vector Machine, with CatBoost as the meta-model.

Results showed that Bitcoin had a higher annualized return and better risk-adjusted performance than the equal-weight portfolio. Portfolios optimized for specific risk metrics like volatility, VaR, and CVaR showed remarkable performance. The vol portfolio, created to minimize volatility, had the highest Sharpe ratio, indicating excellent risk-adjusted performance. Simulations demonstrated the potential of stacking models and optimized portfolios in improving financial performance and managing risks effectively.

Stacking models and optimized portfolios improved risk-adjusted performance, but the lack of statistical significance suggests caution in generalizing these results. Advanced forecasting models and risk-minimizing strategies can enhance portfolio performance, but investors should know the volatility and risks associated with cryptocurrencies. The study’s findings provide valuable insights into integrating forecast accuracy with practical trading applications and risk management.

Ahmed, W. M. A. (2024). On the robust drivers of cryptocurrency liquidity: the case of Bitcoin. Financial Innovation, 10(1). https://doi.org/10.1186/s40854-023-00598-9 López de Prado, M. (2016). Building Diversified Portfolios that Outperform Out of Sample. The Journal of Portfolio Management, 42(4), 59–69. https://doi.org/10.3905/jpm.2016.42.4.059 Sebastião, H., & Godinho, P. (2021). Forecasting and trading cryptocurrencies with machine learning under changing market conditions. Financial Innovation, 7(1). https://doi.org/10.1186/s40854-020-00217-x