CHAPTER 3
ECONOMETRIC ANALYSIS OF TIME SERIES DATA
3.1 Some Basic Concepts and Definitions
• Picture of the behavior of the time series with no measure of reliability,
e.g., index numbers and exponential smoothing refers to Descriptive
Analysis whereas
• Inferential Analysis refers to forecasting future values and measure of
reliability
• Time series analysis is the analysis of data organized across units of time.
• A time series is a collection of observations made sequentially in time.
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 1
• A time series is a collection of data obtained by observing a response
variable at periodic points in time.
• It is usually contrasted to cross-sectional designs where the data is collected
and organized across a number of similar units at the same time for every
observation
• Examples: daily closing stock prices, weekly interest rates, monthly price
indices and yearly earnings.
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 2
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 3
For the following two conceptual reasons to consider the time series models:
• The classic regression models assume that all causation is instantaneous.
• This is clearly suspect and behaviors are dynamic- they evolve over time.
• Many economists believe that time series models are theoretically and
fundamentally more important than cross-sectional models.
• Since time series models help us how the systems change across time.
The researcher which uses the time series data faces two problems which
do not exist in the cross sectional data:
i. One time series variable can influence another with a time lag;
ii. If the variables are non-stationary, spurious regression may arise.
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 4
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 5
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 6
3.3 Components of a time series
A time series is made up of one or more components of the following:
Trend: Sometimes a time series displayed either upward or downward
movements in the average value of the variable of interest.
– Measures the average change in the variable per unit time. In other
words, it measures the change in the mean level of the time series.
Seasonal variations: An upward and downward movements within year
and follow regular pattern.
– Periodic variations that occur with some degree of regulations
within a year or shorter.
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 7
Cyclical variations:An upward or downward movements in the variable
of interest over a period of time.
– It may has four phases, namely; peak, contradiction, trough and
expansion.
– Recurring up and down movement which are extended over a long
period, (usually 2 years or more)
Irregular variations: Non-systematic upward/downward movements caused
by short term unanticipated and non-recurring factors that may be
attributed to unpredictable influences.
– The residual effect remained after the trend, cyclical and seasonal
components have been removed.
– Thus the residual effect represents the random error component of a
time series.
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 8
– These four components can be combined either in additive or multiplicative
time series models.
– One of the objectives of time series data analysis is to forecast some future
values of the series. Thus the most widely used is the additive model.
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 9
3.4 Properties of Time Series Data
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 10
2) Time series data often have time-dependent moments (e.g. mean, variance,
skewness, kurtosis).
• The mean or variance of many time series increases over time. This is a
property of time series data called nonstationarity.
• If two independent, nonstationary series are regressed on each other, the
chances for finding a spurious relationship are very high (Granger &
Newbold,1974).
• If the variance of a series is not constant over time, we can model this
heteroskedasticity using models like ARCH, GARCH and others
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 11
3) The sequential nature of time series data allows for forecasting of future
events.
4) Events in a time series can cause structural breaks in the data series. We can
estimate these changes with intervention analysis, transfer function models,
regime switching/Markov models, etc.
5) Many time series are in an equilibrium relationship over time, what we call
cointegration. We can model this relationship with error correction models
(ECM).
6) Many time series data are endogenously related, which we can model with
multi-equation time series approaches, such as vector autoregression (VAR).
7)The effect of independent variables on a dependent variable can vary over
time. We can estimate these dynamic effects with time varying parameter
models.
2-Nov-24 Prepared by Urgaia Rissa(PhD) for MBA Students
12
The reason why we do not estimate time series with OLS are as follows:
• OLS estimates are sensitive to outliers.
• OLS attempts to minimize the sum of squares for errors; time series with a
trend will result in OLS placing greater weight on the first and last
observations.
• OLS treats the regression relationship as deterministic, whereas time series
have many stochastic trends.
• We can do better modeling dynamics than treating them as a nuisance.
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 13
3.5
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 14
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 15
• Shocks to a stationary series are temporary; the series reverts to its long run
mean.
• For non-stationary series, shocks result in permanent moves away from the
long run mean of the series.
• Stationary series have a finite variance that is time invariant; however,
nonstationary series have infinite variance, σ2 → ∞ as t → ∞.
16
2-Nov-24 Prepared by Urgaia Rissa(PhD) for MBA Students
3.5.1 Trend Stationary vs. Difference Stationary
Difference Stationary: It has time series operators, difference operator and
lag operator
– In the difference operator, there are first difference, second difference
and higher order difference operator
– In the lag operator (backward shift operator), there are transforms an
observation of a time series to the previous one.
2-Nov-24 Prepared by Urgaia Rissa(PhD) for MBA Students 17
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 18
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 19
3.5.2
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 20
3.5.3 The Time Series Unit Root Tests
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 21
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 22
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 23
Dickey-Fuller(D-F) Unit Root Test
– The usual t-statistic is not valid, thus D-F developed appropriate critical
values.
– You can include a constant, trend, or both in the estimates regression
tests
– If you accept the null hypothesis, you conclude that the time series has
a unit root.
– In that case, you should make the first difference becomes stationary
series before proceeding with analysis.
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 24
– We can use the ADF test version if we suspect of autocorrelation in the
residuals.
– This model is the same as the DF test, but includes lags of the residuals too.
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 25
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 26
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 27
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 28
29
2-Nov-24 Prepared by Urgaia Rissa(PhD) for MBA Students
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 30
3.5.4 Non-stationarity and Spurious Correlation
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 31
𝑅2
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 32
3.6 Integrated Time Series
• A time series becomes stationary after differencing it once, it is said to be
integrated of order one, denoted as 1(1).
• If it has to be differenced twice, i.e. difference of difference to make it
stationary, it is said to be integrated of order two, denoted as 1(2).
• If it has to be differenced d times to make it stationary, it is said to be
integrated of order d, denoted as 1(d).
• Therefore the terms "stationary time series" and "time series integrated of
order zero" mean the same thing.
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 33
• The autocorrelations in a correlogram of an 1(0) series decline to zero very
rapidly as the lag increases whereas
• An 1(1) series they decline to zero very slowly as the lag increases and
hence, an 1(1) series is said to have a stochastic trend.
• Most non-stationary economic time series generally do not need to be
differenced more than once or twice.
• To sum up, a non-stationary time series is known as an integrated time
series or a series with stochastic trend.
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 34
3.7 Cointegration Test for Long -run Relationship
2-Nov-24 35
Prepared by Urgaia Rissa(PhD) for MBA Students
• Cointegration is the regression of one non-stationary on one or more non-
stationary time series are not results in spurious regression, there is a long –
run equilibrium relationship among the variables.
• When two or more time series move together in an equilibrium
relationship, it is said to be Cointegration.
• Granger (1983) showed that if two variables are cointegrated, then they
have an error correction representation (ECM)
• Then an ECM is estimated using the lagged residuals from previous step
(et-1) as instruments for the long run equilibrium term.
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 36
• If the unit roots tests reveal non-stationarity, we want to test whether the
time-series are cointegrated, which means that individual variables might
be non-stationary.
• It’s possible for linear combinations of non-stationary variables to be
stationary in the first order difference, we can conclude that Xt and Yt are
cointegrated and the OLS estimates are not spurious
• As Granger notes ‘’ a test for cointegration can be thought of as a per-test
to avoid ‘ spurious regression ‘ situation’’.
• In the context of testing for cointegration, just as the DF and ADF tests,
there are also Engle-Granger (EG) and augmented Enger-Granger (AEG)
tests, which now have been incorporated in several software packages.
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 37
3.8 Vector Autoregressive (VAR) Model Estimation
• VAR is a useful model that allows all variables to be endogenous.
• If you have 3 variables, you have 3 equations, with each variable containing
a certain number of lags in each equation.
• You estimate the system of equations and you can then examine how
variables respond when another variable is shocked above its mean( See
Brandt & Williams, Sage Monograph,2007)
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 38
3.9 Static and Dynamic Econometric Models Estimation
2-Nov-24 39
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 Prepared by Urgaia Rissa(PhD) for MBA Students 40
An autoregressive-distributed lag model of order p in the autoregressive
component and of order q in the distributed lag as ARDL(p,q), or alternatively
ADL(p,q):
One simple method of accommodating dynamic adjustment processes into a
statistical model is to parameterise the model in the form of an error
correction model (ECM).
Such an approach has proved very fruitful in applied econometric research for
the long run equilibrium relationship between Y and X.
The error correction model would be:
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 41
2-Nov-24 Prepared by Urgaia Rissa(PhD) for MBA Students
42
and estimates of the parameters i, i=1,...,5 can be used to give estimates of the
parameters 1, 2, , 1 and 2.
The ECM model is not, of course, the only form of dynamic model used in
econometric analysis.
There are many others such as Partial Adjustment Model which is widely
used.
By virtue of a lagged value of the dependent variable appearing as a regressor,
this is a dynamic regression model.
Note, though, that the partial adjustment model, unlike the ECM does not
imply the existence of the lagged value of the explanatory variable Y as a
regressor.
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 43
3.10 Estimating &Comparing Forecasting the Models
We can compare the models by looking at:
– Significance of AR, MA coefficients
– Compare the fit of the models using the Akaike Information Criterion(AIC)
or Schwartz Bayesian Information Criterion BIC (BIC); choose the model
with the smallest AIC or BIC.
– You could compare them to see which one has the smallest forecasting error.
– Whether residuals of the models are white noise (diagnostic checking)
– The modeling process would involve forecasting of the time series using the
various models you had estimated.
2-Nov-24 44
Prepared by Urgaia Rissa(PhD) for MBA Students
3.11 Time Varying Parameter Models
• Theory might suggest that the effect of Xt on Yt is not constant over time.
• We can check for structural breaks in the data set using Chow or other
tests.
• We can estimate time varying parameters with a variety of models,
including:
– Switching regression/threshold models
– Rolling regression models
– Kalman filter models
– Random coefficients model
2-Nov-24 45
Prepared by Urgaia Rissa(PhD) for MBA Students
Example, in the following data, check non-stationarity and cointegration
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 46
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 47
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 48
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 49
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 50
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 51
• The next step is to check if GDP and consumption are cointegrated, that is
even if both series are non-stationary in level
• There may exist a linear combination of GDP and consumption that is
stationary integrated order of zero, (I(0)).
• Because of the same integrated order of the variable, we apply the Johansen
approach for cointegration test.
• If such is the case, we will use this linear combination, called error
correction term(ECT) as E-Views output was displayed below.
2-Nov-24 Prepared by Urgaia Rissa(PhD) for MBA Students
52
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 53
Normalized Cointegrating Coefficients: 1 Cointegrating Equation(s)
Consumption GDP CONSTANT
1.00000 -0.3470 95.2861
(0.0068)
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 54
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 55
Prepared by Urgaia Rissa(PhD) for MBA Students
2-Nov-24 56