The Research Department of Economics and Management of Tadbir Nikan Business, Marketing, and Finance Open 3092-6238 In Press 2027 07 01 A Financial Engineering Approach to Cryptocurrency Portfolio Risk Management Using Copula-GARCH Models, Extreme Value Theory, and Machine Learning A Financial Engineering Approach to Cryptocurrency Portfolio Risk Management Using Copula-GARCH Models, Extreme Value Theory, and Machine Learning 1 18 EN Journal Article 2026 03 22 <p>This study aimed to develop and evaluate an integrated financial engineering framework for cryptocurrency portfolio risk management among Iranian investors by combining Copula-GARCH models, Extreme Value Theory, and machine-learning techniques. This applied quantitative study used a longitudinal financial time-series design based on daily data from ten major cryptocurrencies, including Bitcoin, Ethereum, Binance Coin, Ripple, Cardano, Dogecoin, Litecoin, Stellar, Tron, and Bitcoin Cash, from January 1, 2019, to December 31, 2024. After converting synchronized daily closing prices into logarithmic returns, the final dataset included 21,910 asset-return observations. Data were analyzed using descriptive diagnostics, Jarque-Bera normality tests, ARCH-LM tests, GARCH-family volatility models, copula dependence structures, Extreme Value Theory, Value at Risk, Conditional Value at Risk, portfolio optimization, machine-learning forecasting models, and VaR backtesting procedures. The inferential results showed significant non-normality and conditional heteroskedasticity in all cryptocurrency return series, as indicated by Jarque-Bera and ARCH-LM tests (p &lt; 0.001). GARCH-family estimations confirmed strong volatility persistence across all assets, with GJR-GARCH and EGARCH models outperforming standard specifications for most cryptocurrencies. Copula-GARCH and EVT results showed that the equal-weighted portfolio had the highest tail-risk exposure, whereas the machine-learning-enhanced CVaR portfolio produced the most favorable risk profile. In VaR backtesting, the variance-covariance and historical simulation models failed the Kupiec and Christoffersen coverage tests, while the Copula-GARCH-EVT-ML model produced the most accurate 99% VaR performance, with observed violations matching expected violations and non-significant coverage test results. Among machine-learning models, Extreme Gradient Boosting achieved the lowest prediction error and strongest directional accuracy. The findings indicate that cryptocurrency portfolio risk cannot be adequately managed through static correlation-based or variance-based models. Integrating Copula-GARCH, Extreme Value Theory, and machine learning provides a more robust framework for estimating volatility, nonlinear dependence, extreme downside risk, and optimized portfolio allocation in cryptocurrency markets.</p> https://www.bmfopen.com/index.php/bmfopen/article/download/502/359