Time Series: Some Basic Concepts

M Ramachandran
Department of Economics
Pondicherry University
 What is a Stochastic Process?

A collection of random variable ordered on time

Let y be a random variable
If it is continuous y (t )
If it is discrete yt
Hence, Most of the economic time series are discrete
random variable. Each observation is a realization of all the
possible outcome
For instance, GDP of Rs.200 billion for 2007 is a realization of
all the possible outcome
What is white noise process ?

E ( yt ) =  = 0

E ( yt −  ) =
2 2

yt =  i yt −i + u t
i = 0 i = 1,2,3,...., n
 White noise process

yt = u t

-1

-2

-3

-4
1 26 51 76 101 126 151 176 201 226 251 276 301 326 351 376 401 426 451 476
What is a Stationary Stochastic process?

yt is a stationary process if its mean & variance and

autocovariance are time invariant

E ( yt ) = 
E ( yt −  ) 2 = 2
E ( yt −  )( yt +k −  ) = rk

If k = 0 then r =  2
Variable yt is a stationary process if

yt = yt −1 + ut ; where   1

y1 =  y0 + u1

y2 =  y1 + u2

=  ( y0 + u1 ) + u2

=  2 y0 + u1 + u2
t −1

y t =  t y0 +   i u t −i
i =0
 The mean of yt

E ( yt ) =  t y0

The variance of yt

E [ yt −  t y0 ] 2 = E [(  t +  t −1 +  2 t −2 + ...)2 ]

=  2 (1 +  2 +  4 + ...)

2
=
1−  2
 Stationary stochastic process
yt = 0.25 yt −1 + ut
4
3
2
1
0
-1
-2
-3
-4
1 51 101 151 201 251 301 351 401 451

yt = 0.75 yt −1 + ut
6

-2

-4

-6
1 51 101 151 201 251 301 351 401 451
What is a non-stationary stochastic process?
A variable y t is non-stationary, if it has time varying
mean or time varying variance or both
There are three types.
1. Random walk without a drift
yt = yt −1 + ut
where ut is a white noise process
y1 = y0 + u1
y2 = y1 + u 2
= y0 + u1 + u2
yt = y0 +  ut
 E ( yt ) = y0
E ( y t − y 0 ) 2 = t . 2

Which is a long memory series

Note that yt = ut is stationary process

 Random walk process
yt = yt −1 + ut
15

10

0
1 51 101 151 201 251 301 351 401 451

-5

-10

-15

-20
2. Random walk with a drift
yt =  0 + yt −1 + ut

y1 =  0 + y0 + u1
y2 =  0 + y1 + u2
=  0 +  0 + y0 + u1 + u2
= 2 0 + y0 + u1 + u2

yt = t 0 + y0 +  ut
 E ( yt ) = t 0 + y0
E ( yt − t 0 − y0 ) 2 = t 2

Hence, both mean and variance are time varying

yt =  0 + ut

If 0  0 yt trends downward
If 0  0 yt trends upward
 Random walk process with a drift
yt = 0.5 + yt −1 + ut
300

250

200

150

100

50

0
1 51 101 151 201 251 301 351 401 451

yt = −0.5 + yt −1 + ut
0

-50

-100

-150

-200

-250

-300
1 51 101 151 201 251 301 351 401 451
Random walk with a drift and a deterministic trend

yt =  0 +  1t +  2 yt −1 + ut
0  0 1  0 2 = 1

yt = 0.5 + 0.3t + yt −1 + ut

40000
35000
30000
25000
20000
15000
10000
5000
0
1 51 101 151 201 251 301 351 401 451
 yt = 0.5 − 0.3t + yt −1 + ut

5000

0
1 51 101 151 201 251 301 351 401 451
-5000

-10000

-15000

-20000

-25000

-30000

-35000

-40000
 Random walk process with a drift and deterministic trend
Yt=β0+β1t +Yt-1+ut
Y1= β0+β1 +Y0+u1
Y2= β0+2β1+ β0+β1 +Y0+u1+u2
=2 β0+3β1+Y0+u1+u2
Y3=β0+3β1+Y2+u3
= β0+3β1+ 2β0+3β1 +Y0+u1+u2+u3
=3 β0+6β1+Y0+u1+u2+u3
. . .
. . . Please note that the coefficients of
. . . β1 forms the series 1,3,6,10… such
𝑡(𝑡+1)
that the t-th term is given by 2
𝒕𝟐 +𝒕
Y t= t.β0+ 𝟐 .β1+σ𝒕𝒊=𝟏 𝒖𝒊
Difference Stationary (DS) & Trend Stationary process (TS)

If trend in a time series is predictable then the trend is deterministic. If not

trend is Stochastic

Let yt follows
yt = β0 + β1 t + β2 yt-1 + ut
Pure random walk implies
β0 = 0; β1 = 0; β2 = 1

yt = yt −1 + u t or ∆yt = ut

Therefore, yt is non-stationary, but ∆yt is stationary; hence yt is DS

process
Random walk with drift needs

β0 ≠ 0; β1 = 0; β2 = 1
∆yt = β0 + ut

yt is again DS process
Suppose
β0 ≠ 0; β1 ≠ 0; β2 = 0
yt = β0 + β1t + ut

This is called TS process. The mean of yt is not a constant,

but predictable and the variance is a constant.
 Trend stationary process
yt = 0.5 + 0.3t + ut

160

140

120

100

80

60

40

20

0
1 51 101 151 201 251 301 351 401 451
If subtract mean
yt - β0 - β1t = ut
the resulting series is stationary.

Suppose
β0 ≠ 0; β1 ≠ 0; β2 = 1
yt = β0 + β1t +yt-1+ ut
or
∆yt = β0 + β1t + ut
which means that yt is non-stationary
 What is an Integrated Stochastic Process?
The RWP implies non-stationarity but its first difference is stationary

yt = yt-1+ ut
∆yt = ut
hence, yt is an integrated series or it is integrated of order 1, written as yt~
I(1).
If yt is non-stationary and only ∆2yt is stationary then yt~ I(2) process.

(1 − L) 2 yt = yt + yt −2 − 2 yt −1 = ut
yt − yt −1 = ut

Therefore a stationary series (ut) is an I(0) process.

 Integrated series of order 2
yt = 2 yt −1 − yt −2 + ut
2000

1000

0
1 51 101 151 201 251 301 351 401 451
-1000

-2000

-3000

-4000

5000

4000

3000

2000

1000

0
1 51 101 151 201 251 301 351 401 451
-1000
What is unit root in time series?
Let y t follows
y t = y t − 1 + u t −1    1
if  =1 then yt follows a RWP; hence yt is non-stationary

It is called unit root because  =1

For instance
yt − yt −1 = u t
Can be written as
(1 − L ) yt = ut

If we set (1 − L) = 0 then L = 1
if |  | 1 then yt is stationary
 Unit Root Worksheet
a b c #VALUE! #VALUE! #VALUE!
-1 -1 1 2.236068 -1.61803 0.618034
0.5 -1.5 1 0.5 2 1
-0.2 0.5 1 1.024695 -1.31174 3.811738
-0.3 0.7 1 1.3 -1 3.333333
-0.4 0.7 1 1.445683 -0.9321 2.682104
 Testing for Unit root
The DF test

y t = y t − 1 + u t
Testing for unit root in yt involves testing the null

H0 :  = 1 against an alternative

H1 :   1
However, the best way to handle this problem is to manipulate the equation
yt = yt −1 + ut
H 0 :  = 0 ( =  − 1)
Hence, the null is yt has a unit root
 The ADF test

m
yt = yt −1 +   y
i =1
i t −i + ut

H0 : = 0 ; H1 :   0
THANK YOU