Part II
METHODS FOR TESTING UNIT ROOTS
Augmented Dickey-Fuller Test (ADF)
Extensions to the Dickey-Fuller Test (DF) e.g “Pantula principle”, seasonal unit roots, structural breaks, fractional unit roots etc.
Phillips-Perron Test (PP)
Kwaitkowski-Phillips-Schmidt-Shin (KPSS)
Elliot-Rothenberg-Stock point optimal (ERS)
Ng-Perron test
Etc.
We only focus on the ADF.
In all cases the basic idea is to search for the data generating process (DGP) from;
Pure random walk
Random walk with drift
Random walk with drift and time trend
3. THE DICKEY FULLER TEST
3.1 Pure random walk
Where
Subtracting on both sides
Let
Then
The DGP is non-stationary if . The DGP is stationary if . Recall the Monte Carlo simulations
This parameterisation facilitates the use of the t-test.
H0: - Non-stationarity
H1: – Stationarity
3.2 Random walk with drift
(i) Test for stationarity alone
H0: - Non-stationarity
H1: - Stationary
(ii) Identifying joint significance of unit root and the drift term
The main question here is whether nature generated the variable as a random walk with drift or not.
This should be done by the following joint test
H0: - Non-stationarity
Failure to reject, whilst rejecting H0: while rejecting H0: and that follows an asymptotic normal distribution.
3.3 Random walk with drift and trend
(i) Test for stationarity alone
For a DGP for random walk with drift, insert a constant/drift term (m)
H0: - Non-stationarity
H1: - Stationary
follows distribution
(ii) Identifying joint significance of unit root and the time trend
The main question here is whether nature generated the variable as a random walk with time trend or not.
This should be done by the following joint test
H0: - Non-stationarity
Failure to reject H0: while rejecting H0: and that follows an asymptotic normal distribution.
Share with your friends: |