Abstarct: This study aims to forecast the volatility of the Iraqi stock market using deep learning approaches, including Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and a hybrid CNN-LSTM model.Given the nonlinear, dynamic, and time-dependent nature of financial markets, deep learning models are considered powerful tools for time series analysis and forecasting.In this study, macroeconomic and market-related variables affecting the Iraqi stock market were collected, preprocessed, normalized, and transformed into a time-series structure using a sliding window approach for model training and testing.The empirical results indicate that all three models are capable of capturing volatility patterns in the Iraqi stock market; however, the CNN-LSTM hybrid model demonstrates superior performance compared to standalone LSTM and GRU models.By combining convolutional laxyers for local feature extraction and LSTM laxyers for learning long-term temporal dependencies, the hybrid model provides more accurate and stable predictions under volatile market conditions.In contrast, although LSTM and GRU models also show acceptable performance, their predictive accuracy decreases during periods of high market fluctuation. Furthermore, the findings reveal that macroeconomic variables such as crude oil prices, exchange rates, and the overall market index play a significant role in explaining stock market volatility in Iraq.This reflects the oil-dependent structure of the Iraqi economy and highlights its sensitivity to external shocks.In addition, trading volume and transaction value contribute as complementary internal factors in explaining market behavior. Overall, the results suggest that the Iraqi stock market exhibits nonlinear and time-dependent behavior, and deep learning models are effective tools for forecasting such complex financial time series.Moreover, hybrid architectures can significantly enhance prediction accuracy and provide deeper insights into financial market dynamics.