Multi-task CNN

My aim was to predict stock prices using a deep learning approach. To achieve this, I developed a Convolutional Neural Network (CNN) model that integrates Long Short-Term Memory (LSTM) and convolutional filtering. Then, I enhanced the model through hyperparameter optimization and data normalization, and incorporated embedded coding, an attention mechanism layer, and dummy variables into a multi-task learning framework to improve the information coefficient (IC). The achievement of this project is the generation of alpha factors, conducting factor backtesting, and calculating key metrics like annualized returns, which reached up to 28%.