Real-Time Sentiment Analysis for Stock Price Forecasting Using ARIMAX Models

Digital news outlets are increasingly shaping financial markets due to the rapid flow of information that is disseminated through them. The stock price forecasting models which have been in use traditionally are mainly dependent on past price information and they do not seem to effectively represent how investors respond to new information. The research questions covered by the study are on whether the use of financial news sentiment in time-series forecasting models can enhance the predictive system in terms of accuracy and relevance of predictions to decisions in the short run when predicting stock prices. The study is based on a quantitative and predictive research design where it uses natural language processing method, namely FinBERT to derive sentiment scores based on U.S. financial news headlines. These sentimental measurements are combined with historical stock price data to come up with hybrid forecasting models. The prevailing price-only models are contrasted with sentiment enhanced models on objective evaluation criteria, such as the Root Mean Squared Error (RMSE) and Mean Absolute error (MAE). Following the Forecasting: Principles and Practice (FPP3) model, the study pays attention to transparency, simplicity, and decision-oriented modeling. The results indicate that the sentiment-enhanced models are capable of doing better than the traditional price-based models in some circumstances, especially when the quality of the data and modeling assumptions are met. The research has a contribution to the scholarly literature and applied financial decision making through illustration on how sentiment-based forecasting could be used to make more informed and justifiable investment decisions.