Diagnostics
Before any forecasting model runs, every series is tested for stationarity, and the group of series is tested for cointegration. These results decide whether the VAR or the VECM model is used.
Stationarity (Augmented Dickey-Fuller test)
A series is stationary if its statistical properties (mean, variance) don't drift over time. Non-stationary series get differenced before modeling. See ADF test in Learn mode.
| Series | ADF statistic | p-value | Verdict | Differencing |
|---|---|---|---|---|
| cpi_all_items | -15.205 | 0.0000 | cpi_all_items is stationary (ADF p-value = 0.0000). | log_difference |
| overnight_rate | -21.226 | 0.0000 | overnight_rate is stationary (ADF p-value = 0.0000). | first_difference_bp |
| spread_10y_2y | -15.497 | 0.0000 | spread_10y_2y is stationary (ADF p-value = 0.0000). | first_difference_bp |
| usdcad | -17.234 | 0.0000 | usdcad is stationary (ADF p-value = 0.0000). | log_difference |
| yield_10y | -80.349 | 0.0000 | yield_10y is stationary (ADF p-value = 0.0000). | first_difference_bp |
| yield_2y | -35.354 | 0.0000 | yield_2y is stationary (ADF p-value = 0.0000). | first_difference_bp |
| yield_5y | -31.172 | 0.0000 | yield_5y is stationary (ADF p-value = 0.0000). | first_difference_bp |
Cointegration (Johansen test)
Tests whether the level series share a stable long-run relationship. See Johansen test in Learn mode.
| Trace statistic | 95% critical value | Cointegration rank | Verdict |
|---|---|---|---|
| 193.833 | 95.754 | 1 | Use VECM |
The trace statistic (193.83) exceeds the 95% critical value (95.75), indicating a stable long-run relationship among the series. Use VECM for forecasting.