Economic Forecasting Model

Section 5: Statistical Validation & Accuracy Assessment

5.1 Validation Philosophy

Validation is treated as a scientific process, not just a metric. Accuracy claims (e.g., 95%) must be metric-specific and time-horizon-specific.

5.2 Evaluation Metrics

  • RMSE (Root Mean Squared Error)
  • MAE (Mean Absolute Error)
  • MAPE (Mean Absolute Percentage Error)
  • R² Score

    • 5.3 Backtesting Strategy

    The system uses rolling-origin backtesting.

  • Train on historical data until time $T$.
  • Predict future window $[T+1, T+H]$.
  • Slide forward to $T+1$ and repeat.

    • 5.4 Confidence Intervals

    Prediction uncertainty is estimated using:

  • Residual analysis
  • Monte Carlo dropout
  • Quantile regression

    • Confidence bands are included in all outputs.

    Previous SectionML Model Design (LSTM/GRU)Next Section Model Versioning & MLOps