A SIMPLE USE TO UNIT ROOT TEST AND ITS APPLICATION USING EVIEWS ·           Definition of   Unit root test ·          Stationary...


·         Definition of  Unit root test

·         Stationary and Non stationary

·         Unit root test statistics

·         Importance of unit root test

·         Application  in Eviews using DGP model

The unit root test is one way to ascertain the stationarity of a data series. It test whether a time series variable is non stationary and possesses unit root. Unit root test enable us to ascertain the order of integration of a series.

Recent advances in econometrics have however shown that oftentimes economic data (mostly time series data) are not as well behaved. Consequently, different data series may not exhibit the same characteristics. It is thus possible to have some data series that exhibit the characteristics of diverging away from their mean with the passage of time, while others may converge to the mean overtime.

Stationary and Non Stationary

   Covariance stationarity of a variable (Z) implies that overtime, Z has ;

-Constant mean

-Constant variance

-Covariance between different observation do not depend on time (t), only on the distance on lag between them (j)

 Cov[Zt, Zt+j]=Cov[Zs, Zs+j]

A time series with the characteristics is known as weakly or covariance stationary. This means the order of integration is at level or I(0).

Z is said to be non stationary if any of this condition do not hold. We can say that data series that diverge from their mean are said to be non stationary. A series is said to be non stationary when it possesses a unit root. The divergence away can be in one or two possible directions; an upward direction and a downward direction, in any case of which the data series is said to exhibit a trend. To utilize such data in any meaningful regression analysis, then we proceed by purging the trend. This is more technically referred to as detrending of the data series.

Detrending of data can be accomplished using one of two possible ways;

·         Incorporating an explicit trend variable in the model specified. A process with deterministic trend is stationary but not a unit root and it is called Trend stationary process (TSP). In TSP, stationary is achieve via: the explicit inclusion of trend variable. For example

        Zt= a0+a1t+Ut      [1]

    t is a time trend and Ut is the error term. After running a regression, we obtain.

     Ũ=Zt - b0 – b1t        [2]

Ũ represent the detrended  Zt.

·         Taking successive difference of the variable in the model such that, we then run our regression using data on the variables in the specification not in their level form, but in the their first difference form. This is called Difference stationary process (DSP).

    A non stationary process that can be differences to achieve stationary is called INTEGRATED NON STATIONARY SERIES. The number of times the series need to be differenced to achieve stationarity on the other hand defines the order of integrated series. If a series is difference once, i.e I(1), it is integrated of order 1. If it is achieved at differencing twice, it is integrated of order 2 or I(2).

  Where the problem come from is that most time series economic data are non stationary and if we run a non stationary data in a regression, we get a result called SPURIOUS Regression.


1.      Dickey Fuller(DF) Test

Dickey and Fuller have proposed the use of a likelihood ratio (LR) test for unit root. This is basis of the so called Dickey Fuller test 

- Intercept:             ∆Zt= µ+ b Zt +et       [3]

-intercept and time trend:            ∆Zt = µ+ bZt + ɑ t +et     [4]

The null hypothesis Ho: b=0; Zt has a unit root. Rejecting the unit root test implies the Zt series is stationary.


If Z is integrated at I(0), then it is stationary at level. No differencing .

If Z is integrated at I(1), then its difference once to be stationary .

If Z is integrated at I(2), then it is the difference of the difference of Z will be stationary.

Note: DF test assumes that the errors are white noises. To get a better test we need to move beyond white noise disturbance and use different version of test allowing for higher order lag.

2.       Augmented Dickey Fuller (ADF) Test

                    ∆Zt = µ +ɑ t + bZt -1+ t-1+et         [5]

The Augmented Dickey Fuller for unit root takes the form of adding the lagged value of a series. et is a pure white noise error, is the maximum length of  lag dependent variable. The essence of the ADF is to improve the statistical fitness of the models. As with  DF, ADF test if the coefficient Zt-1 (b)≠ 0.

The ADF statistics; used in the test, is a negative number. The more negative number it is, the stronger the rejection of the hypothesis that there is a unit root at some level of confidence (Wikipedia)


The H0: b=0; Z has a unit root

If the ADF t-statistics less than any and/or all the t-critical value, we accept the null hypothesis but if its greater than all critical value , we reject the null hypothesis. Note: Putting ADF t-stat  and  all critical value in absolute form.

/t-stat/</t-crit/, we do not reject H0. Therefore, Z is non stationary.

/t-stat/>/t-crit/, we reject the H0. Therefore, Z is stationary.

DF and ADF have been found to have low power in certain circumstances :

-it is difficult to distinguish between b=0.97 and b=1, especially in small sample.

- ADF has a low power in the case of trend stationary process. They fail to reject.

3.      Phillips-Perron (PP) Test

Phillips-Perron  correct for any serial correlation and heteroscedasticity in the errors by directly modifying the test statistics. In Phillips-Perron  test, the lag lengthened not to be specify.

∆Zt= µ*t*t + bZt -1+et       [6]

The PP test builds on the DF of null hypothesis. Like ADF, PP addresses the issue that process generating data for Zt might have a higher order autocorrelation than is admitted in the equation ,making Zt-1 exogenous and thus invalidating the DF test.

Davison and Mackinon (2004) report that the Phillip-Perron test performs worse in finite sample than ADF test.


The H0:b=0; Z has a unit root

If the PP t-statistics less than any and/or all the t-critical value, we accept the null hypothesis but if its greater than all critical value , we reject the null hypothesis. Note: putting  PP t-stat  and  all critical value in absolute form.

/t-stat/</t-crit/, we do not reject H0. Therefore, Z is non stationary.

/t-stat/>/t-crit/, we reject the H0. Therefore, Z is stationary.

4.      Kwiatkowski- Phillips- Schmidt-Shin(KPSS) Test

The KPSS test is a stationary test, the null hypothesis implies that Zt is I(0).

Unlike DF,ADF, and PP that are called unit root test, KPSS is a stationary test


H0: b=0, Z is trend stationary.

H0: Variance et =0. Therefore , mean is constant and Z is trend stationary .

H1: Variance et >0. Therefore , non stationary exist in Z series.

If KPSS t-stat is greater than  all critical value, we reject  the H0 but if its less than all critical value  we accept the H0.

t-stat>t-crit, we reject the H0. Therefore, Z is  non stationary

t-stat <t-crit, we do not reject H0. Therefore, Z is  stationary.


Checking the graph of a series is important before testing for unit root. The graph tells us the correct specification i.e whether  to include intercept or trend and intercept or  none  in the test equation.


Whether a data series has unit root or not, it has a far reaching implications for economic analysis and policy formation and interpretation. Where data series has unit root b=1, any shock to the series is long lasting / permanent. Thus, there will be a cumulative divergence away from the mean or trend of the series. The instability in such series will tend to render any policy formulated and implemented on the basis of the model estimated using such series impotent. This is because underlying any policy formation and implementation is the implicit assumption of stability of the series on which the formulated policies are based.

If the hypothesis that b<1 is upheld, the effect of any shock to the series will fade away progressively overtime. This too has implications for any policy formulated on the basis of model estimated with such series. In the light of the forgoing, the critical point of interest in our analysis is the magnitude b. The closer b is to unity, the stronger chance that shocks to the series will be  long lasting and vice versa. For most practical purposes therefore, what is important in our analysis is the size of b- the autoregressive parameter which bears critically on the duration and effect of a shock.

 Application  in Eviews using DGP model

Make sure the Eviews software is open. See more


      Davidson, Russell; Mackinnon, James,G. (2004).Econometric Theory and Methods. New york: Oxford University Press. P. 623.

      Iganinga, B.O. (2009). Introductory Econometrics: Made Easy. Anitop Books. P.151-158.

    Phillips,P.C.B; Perrron, P.(1988). "Testing for Unit Root in Time Series Regression. Biometrika.75(2);345-346.   



ARTICLES,26,Econometrics,19,economics update,45,MACROECONOMICS,9,MICROECONOMICS,2,
Loaded All Posts Not found any posts VIEW ALL Readmore Reply Cancel reply Delete By Home PAGES POSTS View All RECOMMENDED FOR YOU LABEL ARCHIVE SEARCH ALL POSTS Not found any post match with your request Back Home Sunday Monday Tuesday Wednesday Thursday Friday Saturday Sun Mon Tue Wed Thu Fri Sat January February March April May June July August September October November December Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec just now 1 minute ago $$1$$ minutes ago 1 hour ago $$1$$ hours ago Yesterday $$1$$ days ago $$1$$ weeks ago more than 5 weeks ago Followers Follow THIS CONTENT IS PREMIUM Please share to unlock Copy All Code Select All Code All codes were copied to your clipboard Can not copy the codes / texts, please press [CTRL]+[C] (or CMD+C with Mac) to copy