Stock Price Forecasting with Deep Learning

Abstract

In the world of stock market analysis, the application of deep learning methods has surged, incorporating techniques like long short-term memory (LSTM), recurrent and CNN-based networks, and generative adversarial networks (GANs). This project aligns with this trend, focusing on predicting stock price movements during the crucial last ten minutes of trading, utilizing bid and asking price information. Our approach balances model complexity, inference time, and performance, comparing various models including gradient boosting algorithms, LSTMs, and Transformer based methods. Despite observed limitations in these models, our simple Transformer architectures exhibit promising results, showcasing the effectiveness of minimalistic Transformers in capturing temporal dependencies and complex relationships in sequential stock market data with low inference times.

Intro

In the era of big data, the use of deep learning methods for stock price prediction has witnessed a significant surge in popularity [5]. Notably, various techniques, including Long Short-Term Memory (LSTM), recurrent and CNN-based deep neural networks, as well as Generative Adversarial Networks (GANs), have been employed to forecast both short-term and long-term trends in financial markets [4][7]. Given the intricate nature of stock prediction…