Cross-Country Stock Price Forecasting Using Multimodal Temporal Convolutional Fusion of Multilingual Financial Sentiment and Market Data

Abstract

The nature of stock market forecasting is compounded by the nonlinear, volatile and behaviour based nature of financial markets as a complex challenge. Most of the traditional forecasting methods are based on past facts regarding prices and they do not consider the sentiments of the investors and the availability of multilingual data which plays a major role in market dynamics. In order to overcome these weaknesses, this paper suggests a multimodal deep-learning model of cross-country stock price forecasting that incorporates the predictive power of historical financial data with multilingual sentiment data derived through financial news information. The suggested architecture uses the Temporal Convolutional Networks (TCN) to successfully extract the long-range temporal correlations in both the financial time-series data and the sentiment sequences. Moreover, an adaptive attention-based fusion process is presented to acquire the relative weight of the signals of numerical market indicators and sentiment signals in dynamic market conditions. Transformer-based sentiment models are used to process the multilingual financial news to obtain sentiment polarity, sentiment intensity, and probabilistic sentiment representations that are consistent with the daily market observations. The model is tested using multi-country data of large financial markets, such as the United States, India, the United Kingdom, Germany, and Japan. The experimental outcomes indicate that the suggested framework prevails over benchmark frameworks including LSTM, attention-based LSTM, and sentiment only models. The model has lower forecasting errors both in terms of Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) and directional prediction accuracy. On the whole, the results demonstrate the necessity to combine behavioural sentiment indicators and quantitative financial indicators to increase the strength of the forecasts and generalization in the global markets. The suggested framework offers an interpretation-capable and scalable AI-based system of financial forecasting and intelligent investment decision support systems.