Description

1. Core Concepts: The chapter distinguishes between deterministic trends (linear function of time) and stochastic trends (unit root processes), where shocks have permanent effects.
2. Seasonality: It contrasts deterministic seasonality with stochastic seasonality, using simulations to show that a seasonal random walk (SRW) has nonstationary, drifting subseries, while a stable seasonal AR model does not.
3. Brownian Motion: The foundation for modern unit root theory is introduced via Brownian motion, which describes the random behavior of integrated processes.
4. Trend-Stationary vs. Difference-Stationary: A series like \( Y_t = 0.1 - 0.2t + X_t \) (with stationary \( X_t \)) is trend-stationary, not difference-stationary, because removing the deterministic trend yields stationarity.
5. Unit Root Tests: Formal tests like the Phillips-Perron (PP) and ADF tests are applied to logged U.S. GDP. High p-values (\( \approx 0.97 \)) indicate a failure to reject the null hypothesis of a unit root.
6. Spurious Regression:Regressing a random walk on an independent trend yields a misleadingly high \( R^2 \) (0.985), but the nonstationary residuals reveal the regression is spurious.
7. Integration Order: An \( I(1) \) process becomes stationary after one difference, while an \( I(2) \) process (e.g., a quadratic trend plus noise) requires two differences to become stationary.
8. Cointegration: Even if two series are individually \( I(1) \), a linear combination (e.g., \( X_{t,1} - 2X_{t,2} \)) might be stationary, implying they share a common stochastic trend.
9. Testing for Cointegration: The Engle-Granger test is used on financial spreads (e.g., 3-month vs. 10-year Treasury rates). P-values below 0.05 suggest the rates are cointegrated, moving together in the long run.
10. VECM: The Granger Representation Theorem states that cointegrated systems can be represented as a Vector Error Correction Model (VECM), which models both short-run dynamics and long-run equilibrium adjustment.