Skip to content
To determine if a time series is stationary or has the unit root, three methods can be used: A. The most intuitive way, which is also sufficient in most cases, is to eyeball the ACF (Autocorrelation Function) plot of the time series. The ACF pattern with
a fast decay might imply a stationary series. By testing both the unit root and stationarity, the analyst should be able to have a better understanding about the data nature of a specific time series. The SAS macro below is a convenient wrapper of stationarity tests for many time series in the production environment. (Please note that this macro only works for SAS 9.2 or above.)
B. Statistical tests for Unit Roots, e.g. ADF (Augmented Dickey–Fuller) or PP (Phillips–Perron) test, could be employed as well. With the Null Hypothesis of Unit Root, a statistically significant outcome might suggest a stationary series.
C. In addition to the aforementioned tests for Unit Roots, statistical tests for stationarity, e.g. KPSS (Kwiatkowski–Phillips–Schmidt–Shin) test, might be an useful complement as well. With the Null
Hypothesis of stationarity, a statistically insignificant outcome might suggest a stationary series.
$\begingroup$
The SAS System 14:11 Thursday, October 6, 2013 1 The ARIMA Procedure Name of Variable = ln_G_S_Index Period(s) of Differencing 1 Mean of Working Series 0.094293 Standard Deviation 0.316757 Number of Observations 15 Observation(s) eliminated by differencing 1 Autocorrelations Lag Covariance Correlation -1 9 8 7 6 5 4 3 2 1 0 1 2 3 4 5 6 7 8 9 1 Std Error 0 0.100335 1.00000 | |********************| 0 1 0.0026693 0.02660 | . |* . | 0.258199 2 -0.018517 -.18456 | . ****| . | 0.258382 3 0.029440 0.29342 | . |****** . | 0.267025 "." marks two standard errors Inverse Autocorrelations Lag Correlation -1 9 8 7 6 5 4 3 2 1 0 1 2 3 4 5 6 7 8 9 1 1 -0.14763 | . ***| . | 2 0.19526 | . |**** . | 3 -0.27516 | . ******| . | Partial Autocorrelations Lag Correlation -1 9 8 7 6 5 4 3 2 1 0 1 2 3 4 5 6 7 8 9 1 1 0.02660 | . |* . | 2 -0.18539 | . ****| . | 3 0.31522 | . |****** . | Phillips-Perron Unit Root Tests Type Lags Rho Pr < Rho Tau Pr < Tau Zero Mean 0 -11.6883 0.0066 -3.23 0.0033 1 -11.4504 0.0074 -3.23 0.0034 Single Mean 0 -13.7527 0.0129 -3.71 0.0171 1 -12.6667 0.0218 -3.76 0.0157 Trend 0 -14.5288 0.0601 -3.25 0.1144 1 -13.1531 0.1022 -3.20 0.1239 The SAS System 14:11 Thursday, October 6, 2013 2 The ARIMA Procedure Name of Variable = ln_G_S_Index Period(s) of Differencing 1 Mean of Working Series 0.094293 Standard Deviation 0.316757 Number of Observations 15 Observation(s) eliminated by differencing 1 Autocorrelations Lag Covariance Correlation -1 9 8 7 6 5 4 3 2 1 0 1 2 3 4 5 6 7 8 9 1 Std Error 0 0.100335 1.00000 | |********************| 0 1 0.0026693 0.02660 | . |* . | 0.258199 2 -0.018517 -.18456 | . ****| . | 0.258382 3 0.029440 0.29342 | . |****** . | 0.267025 "." marks two standard errors Inverse Autocorrelations Lag Correlation -1 9 8 7 6 5 4 3 2 1 0 1 2 3 4 5 6 7 8 9 1 1 -0.14763 | . ***| . | 2 0.19526 | . |**** . | 3 -0.27516 | . ******| . | Partial Autocorrelations Lag Correlation -1 9 8 7 6 5 4 3 2 1 0 1 2 3 4 5 6 7 8 9 1 1 0.02660 | . |* . | 2 -0.18539 | . ****| . | 3 0.31522 | . |****** . | Augmented Dickey-Fuller Unit Root Tests Type Lags Rho Pr < Rho Tau Pr < Tau F Pr > F Zero Mean 0 -11.6883 0.0066 -3.23 0.0033 1 -12.4302 0.0041 -2.42 0.0197 Single Mean 0 -13.7527 0.0129 -3.71 0.0171 6.91 0.0157 1 -25.2133 <.0001 -3.63 0.0214 6.59 0.0206 Trend 0 -14.5288 0.0601 -3.25 0.1144 6.44 0.0799 1 -45.0252 <.0001 -3.20 0.1265 6.92 0.0622 The SAS System 14:11 Thursday, October 6, 2013 3
chl
51.7k19 gold badges210 silver badges370 bronze badges
asked Oct 28, 2013 at 5:34
$\endgroup$
3
$\begingroup$
Look at the ADF Unit Root Test section.
If your data is a random walk with drift, then it will be under the type 'Single Mean'.
For the ADF test, H0: Non-stationary Ha: Stationary
if P-value < 0.05, you reject the null hypo (H0) and conclude that data series is stationary. It should be as you already differenced the data once.
Under 'Pr < Rho' which stands for the P-value of your Rho (autocorrelation), it is 0.0129 and <0.0001 thus, we reject the null hypo and conclude that the data is stationary.
answered Nov 4, 2013 at 3:44
$\endgroup$
1