Integrating Intelligent and Classical Methods for Prediction of Gold Price using Data Fusion of Sentiment, Economic and Financial Data
Abstract
Given the importance of gold as a store of value and its vulnerability to a spectrum of market, economic, and geopolitical variables, forecasting gold values is a crucial task in the world of financial marketplace. Since gold is influenced by social, political, and economic elements, precise gold value prediction is crucial for making well-informed financial decisions. This involves identifying intelligent and classical econometric methods to generally forecast the daily gold prices. The multisource dataset comprises sentiment from financial news and social media, macroeconomic variables, and financial market indicators. A fine comparison is done between ARIMA, VAR, Prophet, LSTM, XGBoost, and ensemble-based hybrid methods. A data fusion strategy is implemented, combining to increase the predictive power, both early fusing (feature-level) as well as final fusion (decision-level stacking) are used. Results from the daily data from years 2012 to 2024 show that models built through hybrid data fusion outperform single methods, causing forecast error reduction and improvement in directional accuracy. Statistical tests confirm the significance of gains, especially in times of high volatility. Results demonstrate the value of heterogeneous information constructs in predicting the dual characteristics of gold as an asset of wealth and commodity.