VTG-TLN: Volatility Trend Grouping-Temporal Learning Network for Live Stock Market Index Closing Price Forecasting

Abstract

Stock market prediction is a key research area because financial markets are highly dynamic and influenced by corporate performance, economic conditions, market sentiments and global events. However, accurately forecasting live stock market index closing prices is difficult because financial time-series data have short-term and long-term trend dependencies with different temporal characteristics. In this research, Volatility Trend Grouping-Temporal Learning Network (VTG-TLN) is proposed for live stock market index closing price forecasting. In the volatility branch, the proposed VTG-TLN employs a Temporal Convolutional Network (TCN) to capture rapid price fluctuations and short-term dependencies, whereas Long Short-Term Memory (LSTM) learns long-term dependencies and persistent patterns in the trend branch. By integrating these representations, VTG-TLN captures both long-term price trends and short-term market variations, which leads to accurate live stock market index closing price forecasting. Compared with existing External Trend and Internal Components Analysis (ETICA)-LSTM, proposed VTG-TLN method achieves a lower Mean Absolute Error (MAE) of 22.65 and 77.87 on S&P 500 and NASDAQ datasets.