Time Series Analysis
Definition
A Time Series {xt : t T} is a collection of random variables usually parameterized by
1) the real line T = R= (-∞, ∞)2) the non-negative real line T = R+ = [0, ∞)3) the integers T = Z = {…,-2, -1, 0, 1, 2, …}4) the non-negative integers T = Z+ = {0, 1, 2, …}
If xt is a vector, the collection of random vectors {xt : t T}
is a multivariate time series or multi-channel time series.
If t is a vector, the collection of random variables {xt : t T} is a multidimensional “time” series or spatial series.
(with T = Rk= k-dimensional Euclidean space or a k-dimensional lattice.)
Example of spatial time series
The project• Buoys are located in a grid across the Pacific
ocean• Measuring
– Surface temperature– Wind speed (two components)– Other measurements
The data is being collected almost continuouslyThe purpose is to study El Nino
Technical Note:The probability measure of a time series is defined by specifying the joint distribution (in a consistent manner) of all finite subsets of {xt : t T}.
i.e. marginal distributions of subsets of random variables computed from the joint density of a complete set of variables should agree with the distribution assigned to the subset of variables.
The time series is Normal if all finite subsets of {xt : t T} have a multivariate normal distribution.
Similar statements are true for multi-channel time series and multidimensional time series.
Definition:(t) = mean value function of {xt : t T} = E[xt]for t T.
(t,s) = covariance function of {xt : t T} = E[(xt - (t))(xs - (s))] for t,s T.
For multichannel time series(t) = mean vector function of {xt : t T} = E[xt]for t T and(t,s) = covariance matrix function of {xt : t T}
= E[(xt - (t))(xs - (s))′] for t,s T.
The ith element of the k × 1 vector (t) i(t) =E[xit]
is the mean value function of the time series {xit : t T}
The i,jth element of the k × k matrix (t,s)ij(t,s) =E[(xit - i(t))(xjs - j(s))]
is called the cross-covariance function of the two time series {xit : t T} and {xjt : t T}
Definition:
The time series {xt : t T} is stationary if the joint distribution of xt1, xt2, ... , xtk is the same as the joint distribution of xt1+h ,xt2+h , ... ,xtk+h for all finite subsets t1, t2, ... , tk of Tand all choices of h.
Definition:
The multi-channel time series {xt : t T} isstationary if the joint distribution of xt1, xt2, ... , xtk is the same as the joint distribution of xt1+h , xt2+h , ... , xtk+h for all finite subsets t1, t2, ... , tk of T and all choices of h.
Definition:
The multidimensional time series {xt : t T} is stationary if the joint distribution of xt1
, xt2
, ... , xtkis the same as the joint distribution
of xt1+h ,xt2+h , ... ,xtk+h for all finite subsets t1, t2, ... , tk of T and all choices of h.
Time
The distribution of observations at these points in time
The distribution of observations at these points in timesame as
Stationarity
Some Implication of StationarityIf {xt : t T} is stationary then: 1. The distribution of xt is the same for all t T.2. The joint distribution of xt, xt + h is the same
as the joint distribution of xs, xs + h .
Implication of Stationarity for the mean value function and the covariance function
If {xt : t T} is stationary then for t T.(t) = E[xt] =
and for t,s T. (t,s) = E[(xt - )(xs - )]
= E[(xt+h - )(xs+h - )]= E[(xt-s - )(x0 - )] with h = -s= (t-s)
If the multi-channel time series{xt : t T} isstationary then for t T.
(t) = E[xt] = and for t,s T
(t,s) = (t-s)
Thus for stationary time series the mean value function is constant and the covariance function is only a function of the distance in time (t – s)
If the multidimensional time series {xt : t T} isstationary then for t T.(t) = E[xt] = and for t,s T. (t,s) = E[(xt - )(xs - )]
= (t-s) (called the Covariogram)VariogramV(t,s) = V(t - s) = Var[(xt - xs)] = E[(xt - xs)2]
= Var[xt] + Var[xs] –2Cov[xt,xs]= 2[(0) - (t-s)]
Definition:(t,s) = autocorrelation function of {xt : t T}
= correlation between xt and xs.
for t,s T.
sstt
stxx
xx
st
st
,,,
varvar,cov
If {xt : t T} is stationary then (h) = autocorrelation function of {xt : t T}
= correlation between xt and xt+h.
oh
ooh
xxxx
tht
tht
varvar,cov
Definition:
The time series {xt : t T} is weaklystationary if:
(t) = E[xt] = for all t T. and
(t,s) = (t-s) for all t,s T. or
(t,s) = (t-s) for all t,s T.
Examples
Stationary time series
1. Let X denote a single random variable with mean and standard deviation . In addition X may also be Normal (this condition is not necessary)
Let xt = X for all t T = { …,, -2, -1, 0, 1, 2, …}Then E[xt] = = E[X] for t T and
(h) = E[(xt+h - )(xt - )] = Cov(xt+h,xt )= E[(X - )(X - )] = Var(X) = 2 for all h.
. allfor 1 hohh
Excel file illustrating this time series
2. Suppose {xt : t T} are identically distributed and uncorrelated (independent).T = { …,, -2, -1, 0, 1, 2, …}
Then E[xt] = for t T and(h) = E[(xt+h - )(xt - )]
= Cov(xt+h,xt )
000
hhxVar t
0002
hh
The auto correlation function:
0001
hh
ohh
Comment:
If = 0 then the time series {xt : t T} is called a white noise time series. Thus a white noise time series consist of independent identically distributed random variables with mean 0 and common variance 2
Excel file illustrating this time series
3. Suppose X1, X2, … , Xk and Y1, Y2, … , Yk are independent independent random variables with
sincos1
k
iiiiit tYtXx
222 and iii YEXE 0 ii YEXE
Let 1, 2, … k denote k values in (0,)
For any t T = { …,, -2, -1, 0, 1, 2, …}
2sin2cos1
k
iiiii tYtX
2sin2cos1
k
i ii
ii P
tYP
tX
Excel file illustrating this time series
Then
sincos1
k
iiiiit tYtXExE
tht xxEh
0 sincos1
k
iiiii tYEtXE
sincos1
k
iiiii htYhtXE
sincos1
k
jjjjj tYtX
Hence
coscos1 1
k
i
k
jjiji thtXXEh
thtYY jiji sinsin
sincos thtYX jiji cossin thtXY jiji
sinsincoscos1
2
k
iiiiii thttht
if 0 0,0 since jiYYEXXEYXE jijiji
and 222iii YEXE
Hence using cos(A – B) = cos(A) cos(B) + sin(A) sin(B)
k
iii
k
iiii hthth
1
2
1
2 cos cos
and
k
iiik
jj
k
iii
hwh
hh1
1
2
1
2
coscos
0
k
jj
iiw
1
2
2
where
4. The Moving Average Time series of order q, MA(q)
qtqtttt uuuux 22110
Let 0 =1, 1, 2, … q denote q + 1 numbers.
Let {ut|t T} denote a white noise time series with variance 2.
– independent– mean 0, variance 2.
Let {xt|t T} be defined by the equation.
qtqttt uuuu 2211
Then {xt|t T} is called a Moving Average time series of order q. MA(q)
Excel file illustrating this time series
The mean
qtqtttt uuuuExE 22110
t h th E x x
qtqttt uEuEuEuE 22110
The auto covariance function
qhtqhththt uuuuE 2211
qtqttt uuuu 2211
q
jjtj
q
iihti uuE
00
q
i
q
jjtihtji uuE
0 0
q
i
q
jjtihtji uuE
0 0
qi
qihq
ihii
0
if0
2
. if 0 since jiuuE ji . and 22 iuE
qi
qihhq
ihii
0
if0
2
The autocorrelation function for an MA(q) time series
The autocovariance function for an MA(q) time series
qi
qihhq
ii
hq
ihii
0
if0 0
2
0
5. The Autoregressive Time series of order p, AR(p)
Let 1, 2, … p denote p numbers.
Let {ut|t T} denote a white noise time series with variance 2.
– independent– mean 0, variance 2.
Let {xt|t T} be defined by the equation.
2211 tptpttt uxxxx
Then {xt|t T} is called a Autoregressive time series of order p. AR(p)
Excel file illustrating this time series
Comment:
where {ut|t T} is a white noise time series with variance 2. i.e. 1 = 1 and = 0.
11 ttt uxx
An Autoregressive time series is not necessarily stationary.
Suppose {xt|t T} is an AR(1) time series satisfying the equation:
1 tt ux
121 tttttt uuxuxx
tt uuuux 1210
and is not constant.
ttt uEuEuEuExExE 1210
0xE
tt uVaruVarxVarxVar 10
but
20 txVar
A time series {xt|t T} satisfying the equation:
is called a Random Walk.
1 ttt uxx
Derivation of the mean, autocovariance function and autocorrelation function of a
stationary Autoregressive time series
We use extensively the rules of expectation
is stationary.
Assume that the autoregressive time series {xt|t T} be defined by the equation:
2211 tptpttt uxxxx
Let = E(xt). Then
2211 tptpttt uExExExExE
21 p
1 21 p
p
txE
211
or
The Autocovariance function, (h), of a stationary autoregressive time series {xt|t T}can be determined by using the equation:
2211 tptpttt uxxxx
Thus
1 2Now 1 p
11 tptptt uxxx
The Autocovariance function, (h)
Hence
tht xxEh
thtphtpht xuxxE 11
where
tht xxE 11
thttphtp xuExxE
hphh uxp 11
000
hxuEh
xuEhtt
thtux
Now
0ux t tE u x
1 1t t p t p tE u x x u
211 tpttptt uExuExuE
2
The equations for the autocovariance function of an AR(p) time series
21 10 pp
101 1 pp
212 1 pp
323 1 pp
etc
Or using (-h) = (h)
21 10 pp
101 1 pp
212 1 pp
and 011 ppp
phhh p 11 for h > p
Use the first p + 1 equations to find (0), (1) and (p)
Then use
phhh p 11
for h > pTo compute (h)
The Autoregressive Time series of order p, AR(p)
Let 1, 2, … p denote p numbers.
Let {ut|t T} denote a white noise time series with variance 2.
– independent– mean 0, variance 2.
Let {xt|t T} be defined by the equation.
2211 tptpttt uxxxx
Then {xt|t T} is called a Autoregressive time series of order p. AR(p)
is stationary.
If the autoregressive time series {xt|t T} be defined by the equation:
2211 tptpttt uxxxx
Then
p
txE
211
The Autocovariance function, (h), of a stationary autoregressive time series {xt|t T} be defined by the equation:
2211 tptpttt uxxxx
Satisfy the equations:
21 10 pp
101 1 pp
212 1 pp
and 011 ppp
phhh p 11 for h > p
Yule Walker Equations
The autocovariance function for an AR(p) time series
The mean 1 21t
p
E x
Use the first p + 1 equations (the Yole-Walker Equations) to find (0), (1) and (p)
Then use
phhh p 11
for h > pTo compute (h)
The Autocorrelation function, (h), of a stationary autoregressive time series {xt|t T}:
0
hh
The Yule walker Equations become:
011
2
1 pp
111 1 pp
212 1 pp
and
111 ppp
phhh p 11 for h > p
pp
110
1
2
111 1 pp
212 1 pp
Then
111 ppp
phhh p 11
for h > p
To find (h) and (0): solve for (1), …, (p)
Example
Consider the AR(2) time series:xt = 0.7xt – 1+ 0.2 xt – 2 + 4.1 + ut
where {ut} is a white noise time series with standard deviation = 2.0
White noise ≡ independent, mean zero (normal)
Find , (h), (h)
1 21 1 1
1 22 1 1
To find (h) solve the equations:
or 1 (0.7)1 0.2 1
2 0.7 1 0.2 1
thus 0.7 0.71 0.8751 .2 0.8
2 0.7 0.875 0.2 0.8125
1 21 2h h h
for h > 2
This can be used in sequence to find:
0.7 1 0.2 2h h
3 , 4 , 5 , etc.
resultsh 0 1 2 3 4 5 6 7 8
� �h ) 1.0000 0.8750 0.8125 0.7438 0.6831 0.6269 0.5755 0.5282 0.4849 h
pp
110
1
2
To find (0) use:
= 17.778
or
2
1 2
01 1 2
22.0
1 0.70 0.8750 0.20 0.8125
0h h
To find (h) use:
1 21
4.1 4.1 41
1 0.70 0.20 0.1
To find use:
An explicit formula for (h)
Auto-regressive time series of order p.
Consider solving the difference equation:
011 phhh p
This difference equation can be solved by:Setting up the polynomial
pp xxx 11
prx
rx
rx 111
21
where r1, r2, … , rp are the roots of the polynomial (x).
The difference equation
011 phhh p
has the general solution:
h
pp
hh
rc
rc
rch
111
22
11
where c1, c2, … , cp are determined by using the starting values of the sequence (h).
21
2
1
2
1110
ttt uxx 11
10
and
11 01
hhh 11 1 for h > 1
Example: An AR(1) time series
The difference equation
011 hh Can also be solved by:Setting up the polynomial
xx 11
11
1
1 where1
r
rx
Then a general formula for (h) is:
10 since 1111
11
hh
h
cr
ch
tttt uxxx 2211
10
11 and 21
21 21 hhh
for h > 1
Example: An AR(2) time series
2
11 1
1or
Setting up the polynomial
2211 xxx
2
1 2 1 2 1 2
1 1 11 1 1x x x xr r r r r r
2
22
111 2
4 where
r
2
22
112 2
4 and
r
212
211
1 and 11 :Noterrrr
Then a general formula for (h) is:
hh
rc
rch
22
11
11
For h = 0 and h = 1.
2
2
1
1
2
11 1
1rc
rc
211 cc
Solving for c1 and c2.
Then a general formula for (h) is:
hh
rrrrrrr
rrrrrrrh
22121
212
12121
221 1
111
11
and
2121
221
1 11
rrrrrrc
Solving for c1 and c2.
2121
212
2 11
rrrrrrc
If 04 22
1 21 and rr are real and
hh
rrrrrrr
rrrrrrrh
22121
212
12121
221 1
111
11
is a mixture of two exponentials
If 04 22
1 21 and rr are complex conjugates.
ieRiyxr 1ieRiyxr 2
2 2 1where and tan , tanx xR x yy y
cos sin , cos sini ie i e i
cos , sin2 2
i i i ie e e ei
Some important complex identities
The above identities can be shown using the power series expansions:
2 3 4
12! 3! 4!
u u u ue u
2 4 6
cos 12! 4! 6!u u uu
3 5 7
sin3! 5! 7!u u uu u
Some other trig identities:
1. cos cos cos sin sinu v u v u v
2. cos cos cos sin sinu v u v u v
3. sin sin cos cos sinu v u v u v
4. sin sin cos cos sinu v u v u v
2 25. cos 2 cos sinu u u
6. sin 2 2sin cosu u u
ii
ii
eeRReReR
rrrrrr
11
11
2
22
2121
221
ii
ii
eeRReReR
rrrrrr
1
11
12
22
2121
212
sin212
2
iReRe ii
sin212
2
iReeR ii
h
hiii
h
hiii
Re
iReeR
Re
iReRe
sin21sin21 2
2
2
2
Hence
hh
rrrrrrr
rrrrrrrh
22121
212
12121
221 1
111
11
sin212
11112
iRReeeeR
h
hihihihi
sin1
1sin1sin2
2
RR
hhRh
sin1
sincoscossinsincoscossin2
2
RR
hhhhRh
sin1
sincos1cossin12
22
RR
hRhRh
hR
hRRh cotsin
11cos 2
2
hR
hh tansincos
cot11tan if 2
2
RR
hRhD
cos
hR
hhh tansincos Hence
hRhh
sinsincoscos
cos1
222
tan1cos
sincoscos
1 where
D
a damped cosine wave
Example
Consider the AR(2) time series:xt = 0.7xt – 1+ 0.2 xt – 2 + 4.1 + ut
where {ut} is a white noise time series with standard deviation = 2.0
The correlation function found before using the difference equation:
(h) = 0.7 (h – 1) + 0.2 (h – 2)
h 0 1 2 3 4 5 6 7 8� �h ) 1.0000 0.8750 0.8125 0.7438 0.6831 0.6269 0.5755 0.5282 0.4849 h
Alternatively setting up the polynomial
2 21 21 1 .7 .2x x x x x
1 2
1 1x xr r
221 1 2
12
.7 .7 4 .24where
2 2 .2r
221 1 2
22
.7 .7 4 .24and
2 2 .2r
1.089454
4.58945
Thus
hh
rrrrrrr
rrrrrrrh
22121
212
12121
221 1
111
11
1 2 1 21 22.7156r r r r
2 21 2 2 11 21.8578 and 1 0.85782r r r r
2 21 2 2 1
1 2 1 2 1 2 1 2
1 10.962237 and 0.037763
1 1r r r r
r r r r r r r r
1 10.962237 0.0377631.089454 4.58945
h h
h
Another Example
Consider the AR(2) time series:xt = 0.2xt – 1- 0.5 xt – 2 + 4.1 + ut
where {ut} is a white noise time series with standard deviation = 2.0
The correlation function found before using the difference equation:
(h) = 0.2 (h – 1) - 0.5 (h – 2)
h 0 1 2 3 4 5 6 7 8� �h ) 1.0000 0.8750 0.8125 0.7438 0.6831 0.6269 0.5755 0.5282 0.4849 h
Alternatively setting up the polynomial
2 21 21 1 .2 .5x x x x x
1 2
1 1x xr r
221 1 2
12
.2 .2 4 0.54where
2 2 0.5r
221 1 2
22
.2 .2 4 0.54and
2 2 0.5r
.2 1.96 .2 1.961
i
.2 1.96 .2 1.961
i
Thus
1 .2 1.96 ir i R e
2 .2 1.96 ir i R e
where
2 2 2.2 1.96 2R x y
and0.2tan 0.142857,1.96
xy
1thus tan 0.142857 0.141897
2
2
1 2 1Now tan cot cot .14897 2.333331 2 1
RR
1Thus tan 2.33333 1.165905
2
2.538591cos 0.141897 1.1659052h
h
cosFinally h
D hh
R
2 2Also 1 tan 1 2.3333 2.538591D
hRhD
cos
hR
hhh tansincos Hence
hRhh
sinsincoscos
cos1
222
tan1cos
sincoscos
1 where
D
a damped cosine wave
Conditions for stationarity
Autoregressive Time series of order p, AR(p)
The value of xt increases in magnitude and uteventually becomes negligible.
i.e. 11 ttt uxx
If 1 = 1 and = 0.
The time series {xt|t T} satisfies the equation:
The time series {xt|t T} exhibits deterministic behaviour.
11 tt xx
Let 1, 2, … p denote p numbers.
Let {ut|t T} denote a white noise time series with variance 2.
– independent– mean 0, variance 2.
Let {xt|t T} be defined by the equation.
2211 tptpttt uxxxx
Then {xt|t T} is called a Autoregressive time series of order p. AR(p)
Consider the polynomial
pp xxx 11
prx
rx
rx 111
21
with roots r1, r2 , … , rp
then {xt|t T} is stationary if |ri| > 1 for all i.
If |ri| < 1 for at least one i then {xt|t T} exhibits deterministic behaviour.
If |ri| ≥ 1 and |ri| = 1 for at least one i then {xt|t T} exhibits non-stationary random behaviour.
Special Cases: The AR(1) time
Let {xt|t T} be defined by the equation.
11 ttt uxx
Consider the polynomial
xx 11
1
1rx
with root r1= 1/1
1. {xt|t T} is stationary if |r1| > 1 or |1| < 1 .
2. If |ri| < 1 or |1| > 1 then {xt|t T} exhibits deterministic behaviour.
3. If |ri| = 1 or |1| = 1 then {xt|t T} exhibits non-stationary random behaviour.
Special Cases: The AR(2) time
Let {xt|t T} be defined by the equation.
2211 tttt uxxx
Consider the polynomial
2211 xxx
21
11rx
rx
where r1 and r2 are the roots of (x)1. {xt|t T} is stationary if |r1| > 1 and |r2| > 1 .
2. If |ri| < 1 or |1| > 1 then {xt|t T} exhibits deterministic behaviour.
3. If |ri| ≤ 1 for i = 1,2 and |ri| = 1 for at least on i then {xt|t T} exhibits non-stationary random behaviour.
This is true if 1+2 < 1 , 2 –1 < 1 and 2 > -1.These inequalities define a triangular region for 1 and 2.
Patterns of the ACF and PACF of AR(2) Time SeriesIn the shaded region the roots of the AR operator are complex
h kk
h kkh kk
h kk
12
III
2