Quantitative market intelligence
& macro forecasting.

Behind every visualization lies a rigorous, multi-stage machine learning process designed to filter market noise and isolate genuine predictive signal:

Feature Selection: Before any learning begins, the engine process hundreds of macroeconomic and liquidity indicators. It applies statistical filters—including mRMR (Minimum Redundancy Maximum Relevance) and XGBoost RFE—to strip away irrelevant data and lock onto the few drivers that truly move the market.

Stacked Generalization: We do not rely on a single model. The system employs Stacked Generalization, where multiple distinct models analyze the data independently. Their diverse opinions are aggregated into a consensus forecast, reducing the risk of overfitting to any single market regime.

Metalabeling: A raw "Buy" or "Sell" signal is not enough. A secondary "Metalabeling" model analyzes the primary signal's behavior to estimate its probability of success. This acts as a final quality control filter, sizing positions based on statistical confidence rather than raw direction.

Explainability: Every forecast is auditable. The dashboard utilizes Shapley values to deconstruct the decision-making process, showing you exactly which variables—whether it's Net Liquidity or Volatility—drove the specific prediction for the day.