Investigating Deep Stock Market Forecasting with Sentiment Analysis.

Area of Science:

• * Computational Finance
• * Data Science
• * Machine Learning

Background:

• * Financial time series forecasting commonly integrates sentiment analysis for improved model performance.
• * Deep learning architectures and advanced algorithms are increasingly adopted for their efficiency in financial modeling.
• * Sentiment analysis data, when incorporated into feature space, is assumed to enhance forecasting capacities.

Purpose of the Study:

• * To compare state-of-the-art methods for financial time series forecasting that incorporate sentiment analysis.
• * To evaluate the impact of different feature setups, including sentiment scores and stock prices, on forecasting accuracy.
• * To assess the efficiency of various algorithmic schemes in financial forecasting tasks.

Main Methods:

• * An extensive experimental process was conducted, testing 67 distinct feature setups.
• * Stock closing prices and sentiment scores were utilized as input features.
• * Thirty state-of-the-art algorithmic schemes were applied across two case studies: method comparison and input feature setup comparison.

Main Results:

• * Aggregated results identified a prevalent, high-performing method among those tested.
• * A conditional improvement in model efficiency was observed upon incorporating sentiment analysis feature setups.
• * The impact of sentiment data varied depending on specific forecast time frames and datasets.

Conclusions:

• * The study highlights a superior method for financial time series forecasting incorporating sentiment analysis.
• * Incorporating sentiment analysis can conditionally enhance model efficiency, particularly within certain forecast horizons.
• * Feature engineering with sentiment data presents a valuable, albeit context-dependent, strategy for improving financial forecasting models.