Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 33

An Introduction to Multivariate

Time Series
Without Structural Break
Basic Steps
• Step 1: Plot and visually examine the variables.
Determine their growth rates and time trends.
• Step 2: Determine the order of integration of
each variable
• Step 3: If they turn out to be I(1) then go to
cointrgration
• Step 4: If cointegrated go to causality
Intuitive Understanding

Of the Basic Issues


Random Variables
• Random Variable: A function that assigns (real)
numbers to the results of an experiment. Each
possible outcome of the experiment (i.e. value of the
corresponding random variable) occurs with a certain
probability.
• It is thus a variable whose value is not fixed. It can
take on different values with different probability.
• For example the outcome of a coin toss is a random
variable which can take the value “head” with
probability half and the value “tail” with probability
half. It is a priori impossible to determine which value
it will take.
A Series of Random Variables
• Suppose we have the unemployment data of India
from 1950 to 2010:

• Then each of the above values of u are


random variables in the sense that this particular
value that we see displayed in our excel sheet is a
single realization with a probability of occurrence.
There was a positive probability of other values
occurring (say if there was a depression or all
American companies shifted to India).
Stochastic Process
• A collection of random variables is called a
stochastic (or random) process. Thus the entire
series of variables:

• define a stochastic process ordered in time.


Gaussian White Noise
• Statistically therefore a time series (of, say,
unemployment) is not a series of values of
one variable (called unemployment). It is a
process consisting of a series of random
variables each of which has a distribution
parameterized by a mean and a variance.
• If each of these random variables follow N (0,
σ^2) then the process is called a Gaussian
White Noise Process (GWNP).
Joint Distribution of Random
Variables
• Representing a Stochastic Process: Since all
the data points in a series of values over time
(together constituting a stochastic process)
are themselves random variables with a
distribution, when we analyse the variables
together by arranging them in a single series
(that is, when we analyse the process), the
properties of that series/process mimics the
joint distribution of all the random variables of
the series.
Definition of Joint Distribution
• The probability function defined over a pair of
random variables is called the joint
probability distribution. Consider two random
variables

• The joint probability distribution function of


these two random variables is defined as the
probability that and at the same time
• (or intersection) .
Bypassing Joint Distribution: Short
Cuts
• So to analyse the unemployment data for
India over time, we need to know the joint
distribution of the random variables that
constitutes each data point (corresponding to
time).
• This is almost impossible to do. So instead
what we do is to try and guess about the joint
distribution by looking at some broad
parameters.
Short Cuts (contd.)
• To arrive to a conclusion as to what these
parameters should be we ask the intuitive
question: what does a joint probability
distribution really tell us? The answer is: it is a
precise indicator of how all these random
variables perform as a group.
• So we ask the corresponding question: what
can be the over all indicators of variables
acting together as a group?
Short Cuts (contd.)
• The answers is, look at their:
– Combined mean
– Combined variance
– Combined covarience
• Note that these combined parameters are actually
functions (“averages”) of the corresponding
parameters for each of the random variables. Thus,
for instance, for mean:
• Mean {u1950, u1951,….u2010} = f[Mean{u1950}, Mean{u1951},….Mean{u2010}]
The Role of Time
• Clearly therefore a question may arise as to
what are the individual profiles of these
means, variances and covariance's. for
instance we may ask the question: Are these
individual means, variances and covariance's
all constant or are they functions of time?
Example: Mean=µ0,Var = tσ 2

t Random Variable Mean {ut} Var {ut}

1950 U1950 7.5 .0625 (=1×.0625)

1951 U1951 7.5 .125 (=2×.0625)

… … … …

2010 U2010 7.5 3.8125 (=61×.0625)


Concept of Stationarity
• If a situation such as this arises. That if any one of the
combined mean, variance and covariance are functions of
time then we call the stochastic process consisting of the
unemployment rate of India a (weakly) non stationary
stochastic process. Or simply the time series ut as non
stationary.
• In terms of joint probability distributions the requirement is
more stringent: The cumulative distribution function of the
random variables {u1950, u1951..u2010} is equal to the cumulative
distribution function of say {u1900, u1901..u1950} (that is with a lag
of 50 years). If this is so then the unemployment rate of India
will be called strongly stationary.
Why is The Concept of Stationary
Processes Important ?
We will try and answer this problem
in the context of multivariate time
series analysis
Relationship Between Two Time Series
Variables
• Two types of relationships are possible:
– Intrinsic (true relationship)
– Non-intrinsic (spurious relationships)
• There are two main sources of non-intrinsic or spurious relationship:
– Seasonality
– Trend
• Hence to find relationship between variables we have to remove from
each of them the seasonal and the trend component, otherwise any
relationship that we find between them will be spurious.
• Note a non stationary series is a “trending” process (as the combined
means variance and covariance of the random variables are time
dependent). Hence they should first be de-trended (transformed to
Gaussian White Noise Processes) before any relationship between them
can be thought of.
• Hence stationarity/non-stationarity plays an important role in
determining the relationship between two variables.
Analysis
Types of Trend
• There are two types of trends:
– Deterministic
– Stochastic

• Deterministic trend can be regarded as a natural


phenomenon (“what happens naturally”).

• Stochastic trends can be considered as the


memory of the process.
• Both are Problematic for regressions.
Consequences of Deterministic
Trends
Consequences of Stochastic Trends
Formal Definition of Stationarity
(1)

(2)

(3)

Equations (1) and (2) require the process to have a constant


mean and variance, while (3) requires that the covariance
between any two values from the series ( an autocovariance)
depends only on the time interval between those two values
and not on the point in time . The mean, variance and auto-
covariances are thus required to be independent of time.
Characterization of the Problem
• In order to illustrate the application of the conditions for the
stationary in (1), (2) and (3) we consider the process defined
by
(4)

Where is assumed to define a sequence of independently and


identically distributed (IID) random variables with expected
value zero and variance .The process in 4 is stationary
when is less than one in absolute value ,i.e. .
Theorem 1: If ρ<1, yt satisfies (1)
to (3) and hence is stationary.
Proof (Part 1): E(yt )=0
Proof (Part 2): Var(yt )=σ /(1-ρ )
2 2
Proof (Part 3): Cov(yt, yt-τ )=ρτσ2/(1-
ρ2)
Thus

The fact that , and do not depend on


means that the AR(1) process is indeed stationary when
is less than one in absolute value. QED.
Theorem 2: If ρ = 1(or >1), yt
does not satisfy (1) to (3) and
hence is non-stationary
Proof see Hamilton (1994)
Why is it called “Unit Root”?
We refer to being less than one in absolute value as the
stationary condition.
This condition can be expressed in a different way if we return
to equation 4 and write it in the form

Where ,is a linear function of , the lag operator.


The root of this function(i.e.the solution to ) is given by

,so that the requirement that has absolute value less


than one in absolute value.
Furthermore, has a unit root , i.e.the AR(1) process has a
unit root, if and only if is 1.
Way Out
• If there is deterministic trend then detrend
each variable by taking the error of the
regression of that variable with a constant and
a time trend (which is equivalent to including
a time trend in the actual regression with the
parameters redefined - Exercise).
• If there is stochastic trend then difference
before regressing.
• If there is both: (informal thumb rule) usually
differencing will do the trick.
Correcting Stochastic Trend
Suppose εt is white noise and Xt is random walk.
Tests for Unit Root Processes

You might also like