StockQM: A Cross-Frequency Dataset for Stock Prediction and a New Stock Prediction Model

Kuan-Hsien Liu, Yu-Shiang Lin, Tsung-Jung Liu, Wei-Shen Tai

Abstract: Stock prediction is a highly discussed area, attributed to the inherent noise present in stock data, which makes accurate predictions challenging. Consequently, various methods have been developed to forecast stock price trends and devise effective trading strategies aimed at generating excess returns in the stock market. This article introduces a novel approach by leveraging a cross-frequency dataset that incorporates quarterly financial ratios and monthly revenue data. The methodology entails applying LSTMs with an attention mechanism to analyze the fundamental financial ratios of companies. The study compares the performance of this proposed model against conventional methods such as LSTM without attention, multi-factor regression models, Random Forest Regressor, WGAN-GP, and CNN-LSTM. The evaluation reveals that the Mean Absolute Error of the proposed LSTM with attention model outperforms other methods, suggesting its potential as a new tool for future investment strategies.