Time-Series Forecasting & Seasonality Decomposer (Prophet / ARIMA)

Act as a Staff Quantitative Analyst and Time-Series Forecasting Expert.

Build a forecasting model for:
- Target Variable: [e.g. Daily active users, hourly API server load, monthly recurring revenue]
- Historical Range: [e.g. 2 years of daily data]
- Known Seasonality: [e.g. Weekend dips, Black Friday surges, summer lulls]

Provide:
1. Exploratory Data Decomposition: Python code using `statsmodels` (`seasonal_decompose` or STL) to extract Trend, Seasonality, and Residuals.
2. Stationarity Testing: Augmented Dickey-Fuller (ADF) test interpretation and required differencing orders.
3. Forecasting Implementation: Model using either Prophet (incorporating custom country holidays) or SARIMAX with exogenous regressors.
4. Evaluation Framework: Compute Backtesting MAPE, RMSE, and Coverage of 95% Confidence Intervals.
5. Anomaly Detection: Flag historical outliers that exceed 3 standard deviations from predicted bands.