<![CDATA[This study aims to classify Indonesian ANTAM gold market states using a Decision Tree model built on daily price data from 2010–2024. Market conditions are categorized into three classes: bullish, bearish, and sideways, based on forward returns with an adaptive quantile-based thresholding scheme. The feature set comprises multi-horizon rolling volatility indicators (e.g., std_5, std_10, std_20) and momentum measures (e.g., mom_5, mom_10, mom_20). A time-based split is applied, allocating 80% of observations for training and 20% for testing. Evaluation on the test set yields an accuracy of 0.337 with a macro-F1 of approximately 0.34, indicating limited predictive performance in a three-class setting. Interpretability analysis reveals that std_20 is the most influential feature, followed by std_10 and mom_5, while one-day returns contribute marginally. These findings suggest that aggregated volatility and momentum patterns are more informative than single-day fluctuations for market regime mapping. Overall, the Decision Tree is best positioned as an interpretable baseline for transparent market-state analysis, providing a foundation for future work involving richer features and more robust models.]]>
Copyrights © 2026
