Abstract:
Financial markets are complex, dynamic systems influenced by numerous macroeconomic and microeconomic factors. The ability to predict stock price movements and design effective trading strategies is a long-standing challenge for traders, investors, and researchers. With the advent of machine learning (ML) and deep learning (DL) methodologies, datadriven approaches have gained prominence in forecasting stock trends and optimizing trading strategies. This thesis investigates the application of 1-Dimensional Convolutional Neural Networks (1-D CNNs) and eXtreme Gradient Boosting (XGBoost) in trading strategies within the equity market and the futures & options (F&O) segment. The study leverages a rolling window approach, analyzing 22 key financial features from both Indian and Foreign Market indices to improve predictive performance. The research explores various hyperparameter configurations, filter sizes, activation functions, and rolling window sizes to optimize model performance. Our findings indicate that while 1-D CNNs are effective in capturing sequential dependencies in financial time series data, they often struggle with generalization due to overfitting on training data. On the other hand, XGBoost demonstrates robustness by effectively handling structured financial datasets, although hyperparameter tuning remains a key challenge. Additionally, the Fama-French Three-Factor Model (FF3) is explored as a potential sector rotation strategy, offering insights into factor-based investing in the Indian market. The results suggest that combining deep learning models with ensemble learning techniques could enhance stock market prediction accuracy. Future research can focus on integrating hybrid architectures, cost-sensitive loss functions, and alternative feature engineering strategies to further refine predictive models. This work contributes to the growing literature on AI-driven trading strategies and provides a foundation for deploying automated financial models in real-world market scenarios.
