人大高级计量经济学讲义TimeSeriesEconometricscointegration.ppt

Time Series Econometrics: Cointegration Dr.Yi Jingtao University of Nottingham United Kingdom Types of Economic Time Series Principal Feature of Time Series: Trend (deterministic or stochastic) Three Main Types: Time Series with Deterministic Trend Time Series with Stochastic Trend Time Series with Stationarity (Trend-free) Time Series with Stochastic Trend Time Series with Three Types Stationarity A stochastic process is weakly stationary if: Mean is constant for all t; Variance is constant for all t; Covariance is constant for all t. Stationarity Stationarity Stationarity Autocorrelation Function of Stationary Time Series Unit Roots Differences of non-stationary time series to achieve stationarity. When the first difference of time series Yt is required to achieve stationarity, Yt is said to have one unit root. Yt is integrated of order one. Here, Yt~I(1); DYt=Yt-Yt-1~I(0). No. of unit roots in Yt =No. of times to difference Yt to achieve stationarity. If Yt has d unit roots, it is to say Yt~I(d); Yt is integrated of order d. Unit Roots ADF Test for Unit Roots Spurious Regression Non-stationary time series are highly correlated although there are no true relationship between them. This is because they have strong trend components and they are correlated time trends. Spurious Regression Cointegration Conclusion Time series often have a trend, deterministic trend or stochastic trend. Three types of time series: trend-stationary, unit-root, stationary. Regression relies upon stationary time series. Non-stationary series regression is spurious unless there is cointegration between them. Unit-root time series can achieve cointegration if the residuals are trend-free. Cointegration reflects a long-run relationship between non-stationary time series. * A formal and commonly used test of non-stationary time series Yt is the Augmented Dickey-Fuller (ADF) test. where p is selected to produce white noise residuals. *