Chapter 6

Time series analysis

The concept of stationarity
 Cont’d

• is known as the autocovariance function (since it is the

covariance of Yt with its own previous (or lagged) values).
• When s = 0, the autocovariance function is simply the
variance of Yt.
• The autocovariances are not as such particularly useful
measures of the relationship between Yt and its previous
values since their values depend on the units of
measurement, and hence the values that they take have no
immediate interpretation.
Cont’d
 Cont’d

• Thus, a white noise process has zero mean, constant variance

(independent of t), and zero autocovariances.
• Another way to state this last condition would be to say that
each observation is uncorrelated with all other values in the
sequence.
 Autoregressive processes and stationarity

• An autoregressive model is one where the current value of a

variable Yt depends upon only the values that the variable
took in previous periods plus an error term.
• An autoregressive process of order p, denoted as AR(p) , can
be expressed as:

• Stationarity is a desirable property of an AR model for several

reasons.
 Cont’d

• One important reason is that a non-stationary AR process

exhibits the property that previous values of the white noise
error term will have a non-declining effect on the current
value of Yt as time progresses.
• In contrast, the autocovariances (and hence, the
autocorrelations) of a stationary AR process will decline
eventually as the lag length is increased (the autocorrelation
function (ACF) will decay geometrically to zero).
Cont’d
Cont’d
Cont’d
Cont’d
Integrated processes and differencing
Cont’d
 Cont’d
• This concept can be generalized to consider the case where
the series contains more than one ‘unit root’.
• In such cases, the first difference operator would need to be
applied more than once to induce stationarity.
• If a non-stationary series Yt must be differenced d times
before it becomes stationary, then it is said to be integrated of
order d. This would be written as: Yt ~ I(d) .
• An I(0) series is a stationary series, while an I(1) series
contains one unit root.
• An I(2) series contains two unit roots and so would require
differencing twice to induce stationarity.
• The majority of financial and economic time series contain a
single unit root, although some are stationary and some have
been argued to possibly contain two unit roots.
 Testing for a unit root

• We need some kind of formal hypothesis testing procedure

that answers the question, ‘given the sample data at hand, is
it plausible that the true data generating process for Y
contains one or more unit roots?’
• Consider the regression:
Cont’d
 Cont’d

• In regressions involving trended data, Dickey and Fuller have

shown that the conventional test of significance (that is, the t-
test) that compares the above test statistic with the critical
values from the standard t-table tends to incorrectly reject
the null hypothesis
• As a solution to this problem, they have derived an
appropriate set of critical values for testing the hypothesis
that
• Thus, the critical values for the above test are to be referred
from such Dickey- Fuller tables.
Cont’d
Cont’d
 Cont’d
• We can see that the price of sesame exhibits an increasing trend over
time. This might be an indication that the process is non-stationary. To
determine whether this is so, the augmented Dickey- Fuller test is carried
out. E-views output of the ADF test is shown below:

• The ADF test statistic (0.182840) is greater than the critical values at 1%,
5% and 10% significance levels (or the p-value (Prob.*) is greater than
10%). Thus, the null hypothesis of a unit root can not be rejected. This tells
us that the export price of sesame is a unit root process.
Cont’d
 Cont’d
• The p-value of the ADF test statistic for first differences of the export price
of sesame is less than 1%.
• Thus, we reject the null hypothesis and conclude that the first differenced
series is stationary, that is, the export price of sesame is integrated of
order one or I(1).
• Figure 3 is a time plot of the first differences of the export price of
sesame. We can observe that the series revolves around zero with no
apparent trend.
 Non-stationarity and spurious regression

• There are several reasons why the concept of non-stationarity

is important and why it is essential that variables that are non-
stationary be treated differently from those that are
stationary.
• One of reasons is that the use of non-stationary data can lead
to spurious regressions.
• If two stationary variables are generated as independent
random series, when one of those variables is regressed on
the other, the t-ratio on the slope coefficient would be
expected not to be significantly different from zero, and the
value of R2 would be expected to be very low.
 Cont’d

• This seems obvious since the variables are not related to

one another.
• However, if two variables are trending over time, a
regression of one on the other could have a high R2 even if
the two are totally unrelated.
• So, if standard regression techniques are applied to non-
stationary data, the end result could be a regression that
‘looks’ good under standard measures (significant
coefficient estimates and a high R2 ), but which is really
valueless or has no meaningful economic interpretation.
• Such a model would be termed a ‘spurious regression’.
 Cont’d

• Illustration: The following EViews output pertains to a linear

regression of the United States (US) GDP on the total reserves of
Ethiopia (including gold) from 1961 to 2008 (both in current USD).
• We can see that R2 is large (73%) and the model is adequate as
judged by the F-test (p-value < 0.001).
• Moreover, the t-test indicates that the series of total reserves of
Ethiopia is significant at the 1% level, that is, total reserves of
Ethiopia has a significant influence on US GDP.
• But we know that the total reserves of Ethiopia are in no way
related with US GDP.
• This is clearly a spurious regression.
 Cont’d

• The augmented Dickey- Fuller test is carried out to determine

whether the two series are stationary or not.
• E-views output of the ADF tests is shown below. The p-values
of the ADF test statistics are both greater than 10%.
• Thus, both series are non-stationary and the significant
relationship obtained from the linear regression above would
seem to be a consequence of the underlying trend.
Cont’d
Cont’d
 Cont’d
• The result of a linear regression of the first difference of US GDP on the
first difference of total reserves of Ethiopia is shown below.
• Note that this is not spurious regression since the differenced series are
both stationary.
• We can see that the value of R2 is zero and the F statistic is insignificant
(p-value = 0.999 > 0.05). Thus, there is no relationship between US GDP
and the total reserves of Ethiopia.
 Timeseries Models

1. Univariate Timeseries Analysis

2. Multivariate Timeseries Analysis