Predicting volatility and correlation in financial assets is crucial for assessing financial risk and achieving effective portfolio diversification. However, conventional econometric models used for forecasting volatility and dependence structures have frequently been criticized for their limited predictive performance. Consequently, the adoption of deep learning techniques for volatility and correlation forecasting has emerged as a logical alternative. Nevertheless, although the application of machine learning models in volatility prediction has advanced considerably, these approaches still face challenges in effectively capturing market noise and key stylized facts, such as volatility clustering and leverage effects. These considerations highlight the potential for integrating traditional econometric models with deep learning techniques, particularly convolutional neural networks and long short-term memory networks, an area that still requires further exploration. In this thesis, we investigate the integration of GARCH-type models with a CNN–LSTM hybrid framework to enhance the accuracy of volatility prediction in financial markets. Convolutional neural networks (CNNs) are effective in identifying local spatial patterns in timeseries data, whereas long short-term memory (LSTM) networks are capable of retaining crucial information over extended temporal horizons. In addition, GARCH models are well known for their ability to capture key stylized facts of financial returns, including volatility clustering and excess kurtosis. For GARCH-type model selection, several model adequacy and selection criteria are considered, including the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), maximum log-likelihood values, and the statistical significance of model parameters. Furthermore, the mean-reversion property of volatility dynamics is evaluated under alternative distributional assumptions, such as the normal, Student’s-t, and generalized error distributions. The CNN and LSTM models are subsequently selected due to v their complementary capabilities in volatility prediction. The forecasting performance of the proposed hybrid models is evaluated using loss functions computed from daily out-ofsample test data for the S&P 500 index. Specifically, Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE) are employed as evaluation metrics. The empirical results indicate that the proposed GARCH–CNN–LSTM model outperforms alternative hybrid specifications, achieving up to a 13% improvement in forecasting accuracy over the study period. Furthermore, this study proposes a DCC–EGARCH model integrated with a Long Short- Term Memory (LSTM) network to model multivariate volatility and to obtain more precise estimates of dynamic correlations. The DCC–EGARCH framework effectively captures asymmetric effects in standardized residuals within the conditional covariance structure, while the LSTM network is capable of learning long-term dependencies and complex temporal patterns in sequential data. Consequently, the combined framework is expected to enhance forecasting accuracy for dynamic correlations across stock market indices as well as selected agricultural commodities. In particular, the volatility dynamics and time-varying correlations among the S&P 500, the Shanghai Stock Exchange Composite Index (SHCOMP), the NIFTY 50 of the National Stock Exchange of India, and the S&P/TSX Composite Index of Canada are empirically examined. The experimental results indicate that incorporating LSTM into the DCC–EGARCH framework substantially improves forecasting performance, as evidenced by lower Mean Squared Error (MSE) and Mean Absolute Error (MAE) values. Overall, the findings demonstrate that the proposed DCC–EGARCH–LSTM hybrid model outperforms alternative DCC specifications, highlighting the effectiveness of integrating deep learning techniques to achieve more reliable and accurate predictions of dynamic correlations in financial markets. In the nutshell, we integrated econometric ARCH -type models with hybrid deep learning models for more accurate univariate and multivariate volatility and correlation prediction in selected agricultural commodities and stock market indices