Clustering Market Volatility Modes for CatBoost Gold Trend Signals
Summary
The article describes a unidirectional gold trading workflow that combines K-Means clustering with CatBoost classifiers. Rolling standard deviations of closing prices form both the main model’s features and separate meta-features used to identify market modes. Trade labels mark whether price moves far enough in the chosen direction over a randomly selected future horizon, after accounting for markup. A CatBoost model predicts trade direction, while another classifies market clusters; their predictions are passed to a custom tester and can be exported for use in MetaTrader 5.
The models use shuffled training and validation subsets, with early stopping to limit fitting and validation accuracy or F1 to select iterations. The article reports that model quality is comparable to an earlier approach, but the supplied text gives no numerical performance results or robust out-of-sample evidence. Its example uses closing prices and volatility features, and the author suggests testing alternative features. The random split, custom testing setup and limited description of costs and robustness mean the reported promise should be treated cautiously.
Key ideas
- K-Means clusters rolling volatility features to represent market modes.
- Separate CatBoost classifiers predict trade direction and cluster membership.
- Trade labels reflect a directional price move over a randomly chosen future horizon, including markup.
- Validation subsets and early stopping are used during model fitting.
- The reported comparison lacks detailed numerical evidence and does not establish robustness beyond the described tests.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.