introduction to time series analysishalweb.uc3m.es/esp/personal/personas/jtena/eng/... ·...
TRANSCRIPT
Introduction to Time Series
Analysis
Juan de Dios Tena Horrillo
Office. 10.1.02B
1
Intro
ductio
n to
time series (2
013 )
OUTLINE
Chapter 1. Univariate ARIMA models.
Chapter 2. Model fitting and checking.
Chapter 3. Prediction and model selection.
Chapter 4. Outliers and influential
observations.
Chapter 5. Vector Autoregressive and Vector
Error Correction Models.
Chapter 6. Nonlinear Time Series Modelling.
2
Intro
ductio
n to
time series (2
013 )
TEXTBOOKS
Peña, D., Tiao, G.C. and Tsay, R.S. (2000). A Course in Time Series Analysis. Wiley & sons.
Hamilton, J.D. (1994). Time Series Analysis. Princeton University Press.
Box, G.E.P., Jenkins, G.M. and Reinsel, G. (1994). Time Series Analysis: Forecasting and Control, 3rd. Ed. Prentice-Hall, Englewood Cliffs, NJ.
Enders, W. (2004). Applied Econometric Time Series . Wiley & sons.
Lütkepohl, H. and Krätzig, M. (2004). Applied Time Series Econometrics. Cambridge University Press.
Franses, P.H. and van Dick, D. (2000). Non-linear time series models in empirical finance. Cambridge University Press.
3
Intro
ductio
n to
time series (2
013 )
GRADING
Final exam (70%).
Class participation and empirical project (30%).
4
Intro
ductio
n to
time series (2
013 )
WEB RESOURCES
Global Insight.
Time series data library.
http://datamarket.com/data/list/?q=provider:tsdl
Macroeconomic time series
http://www.fgn.unisg.ch/eurmacro/macrodata/index.html
Instituto Nacional de Estadística
http://www.ine.es
Eurostats
http://epp.eurostat.ec.europa.eu/portal/page/portal/eurostat/home/
5
Intro
ductio
n to
time series (2
013 )
MOTIVATION
Time series: sequence of observations taken at
regular intervals of time
Data in bussines, engineering, enviroment,
medicine, etc.
6
Intro
ductio
n to
time series (2
013 )
MOTIVATION.
7
Intro
ductio
n to
time series (2
013 )
SUICIDES IN SPAIN, TOTAL NUMBER
Source: INE
100
150
200
250
300
350
95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10
SUICIDES
MOTIVATION.
8
Intro
ductio
n to
time series (2
013 )
CO2 EMISSIONS IN THE USA
Source: World Databank
2,800,000
3,200,000
3,600,000
4,000,000
4,400,000
4,800,000
5,200,000
5,600,000
6,000,000
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005
CO2
MOTIVATION.
9
Intro
ductio
n to
time series (2
013 )
INDUSTRIAL PRODUCTION INDEX (SPAIN)
Source: ine
30
40
50
60
70
80
90
100
110
120
1975 1980 1985 1990 1995 2000 2005 2010
IPI
MOTIVATION.
10
Intro
ductio
n to
time series (2
013 )
MOTIVATION.
11
Intro
ductio
n to
time series (2
013 )
OVERNIGHT STAYS IN SPANISH HOTELS
Source: ine
2,000,000
3,000,000
4,000,000
5,000,000
6,000,000
7,000,000
8,000,000
9,000,000
10,000,000
11,000,000
99 00 01 02 03 04 05 06 07 08 09 10 11 12
TURISTS
MOTIVATION.
Intro
ductio
n to
time series (2
013 )
12
CONSUMER PRICE INDEX IN GERMANY AND SPAIN
Source: Eurostats
70
80
90
100
110
120
130
1996 1998 2000 2002 2004 2006 2008 2010 2012
GERMAN_CPI SPANISH_CPI
MOTIVATION.
Intro
ductio
n to
time series (2
013 )
13
CONSUMER PRICE INDEX FOR FOOD AND NON-ALCOHOLIC
BEVERAGES IN GERMANY AND SPAIN
Source: Eurostats
7 0
8 0
9 0
1 0 0
1 1 0
1 2 0
1 9 9 6 1 9 9 8 2 0 0 0 2 0 0 2 2 0 0 4 2 0 0 6 2 0 0 8 2 0 1 0 2 0 1 2
S P A N IS H _ F O O D G E R M A N _ F O O D
MOTIVATION.
Intro
ductio
n to
time series (2
013 )
14
Source: Instituto Flores de Lemus
-3
-2
-1
1
2
3
4
5
6
2008 2009 2010 2011 2012 2013 2014
YEAR-ON-YEAR RATE OF INFLATION IN SPAIN AND CONTRIBUTIONS OF MAIN COMPONENTS
ENERGÍA ALIMENTOS NO ELABORADOS
INFLACIÓN SUBYACENTE IPC TOTAL
MOTIVATION.
Intro
ductio
n to
time series (2
013 )
15
Source: Instituto Flores de Lemus
-2.5
-1.5
-0.5
0.5
1.5
2.5
3.5
4.5
5.5
2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
INFLATION IN MADRIDAnnual rates of growth
80% 60% 40% 20%Confidence intervals:
Average inflation(1998-2011): 2,7%
MOTIVATION.
16
Intro
ductio
n to
time series (2
013 )
MOTIVATION.
17
Intro
ductio
n to
time series (2
013 )
MOTIVATION.
Knowledge of the dynamic structure will help to
produce accurate forecasts of future observations
and design optimal control schemes.
18
Intro
ductio
n to
time series (2
013 )
MOTIVATION
Main objectives
Understanding the dynamic structure of the
observations in a single series.
Ascertain the leading, lagging and feedback
relationships among several variables.
Analyzing the impact of other variables that
can be used for policy decisions.
What to do when there are breaks in the
evolution of the series. 19
Intro
ductio
n to
time series (2
013 )
Chapter 1
Univariate ARIMA Models
Recommended readings: chapters 1 to 3 of Peña et al. (2000)
20
Intro
ductio
n to
time series (2
013 )
CHAPTER 1. CONTENTS.
1.1. Introduction.
Definitions, examples,
1.2. Properties of univariate ARMA.
ARMA weights.
Stationarity and covariance structure.
Autocorrelation function.
Partial autocorrelación function.
1.3. Model especification.
1.4. Examples.
21
Intro
ductio
n to
time series (2
013 )
INTRODUCTION
22
Intro
ductio
n to
time series (2
013 )
Stochastic Process. Real-valued random
variable 𝑍𝑡 that follows a distribution 𝑓𝑡(𝑍𝑡) .
The T-dimensional variable 𝑍𝑡1, 𝑍𝑡2
, … , 𝑍𝑡𝑇
will have a joint distribution that depends on
𝑡1, … , 𝑡𝑇 .
INTRODUCTION
23
Intro
ductio
n to
time series (2
013 )
Time series 𝑧𝑡1, 𝑧𝑡2
, … , 𝑧𝑡𝑇 will denote a
particular realization of the stochastic process.
To simplify notation 𝑧1 , 𝑧2 , … , 𝑧𝑇 .
INTRODUCTION
24
Intro
ductio
n to
time series (2
013 )
Assumptions
1. The process is stationary.
2. The joint distribution of 𝑧1 , 𝑧2 , … , 𝑧𝑇 is a
multivariate normal distribution.
INTRODUCTION
Implications of stationarity
The joint distribution remains constant over time
It is then true that
25
Intro
ductio
n to
time series (2
013 ) ),...,,(),...,,( 2121 kTkkT zzzfzzzf
;ztEz zt VVz
INTRODUCTION
Stationarity implies a constant mean and bounded
deviations from it.
Strong requirement, few actual economic series will
satisfy it.
26
Intro
ductio
n to
time series (2
013 )
INTRODUCTION
27
Intro
ductio
n to
time series (2
013 )
INTRODUCTION
28
Intro
ductio
n to
time series (2
013 )
INTRODUCTION
29
Intro
ductio
n to
time series (2
013 )
INTRODUCTION
Transformations to achieve stationarity
Constant variance: log/level plus outlier correction.
Stationary in mean: differencing.
30
Intro
ductio
n to
time series (2
013 )
31
Intro
ductio
n to
time series (2
013 )
32
Intro
ductio
n to
time series (2
013 )
INTRODUCTION
Differencing
Let B be the backward operator
We shall use the operators:
Regular difference
Seasonal difference
33
Intro
ductio
n to
time series (2
013 )
jttj zzB
B 1s
s B 1
INTRODUCTION
34
Intro
ductio
n to
time series (2
013 )
If 𝑧𝑡 is a deterministic trend (𝑧𝑡 = 𝑎 + 𝑏𝑡)
∇𝑧𝑡 = 𝑏
∇2𝑧𝑡 = 0
where
∇2𝑧𝑡 = ∇(∇𝑧𝑡)
In general, ∇d will reduce a polynomial of degree d
to a constant.
INTRODUCTION
35
Intro
ductio
n to
time series (2
013 )
Example: for a quarterly series 𝑧𝑡 , ∇4𝑧𝑡 = 𝑧𝑡 − 𝑧𝑡−4
will cancel a constant, but it will also cancel other
deterministic periodic functions.
INTRODUCTION
36
Intro
ductio
n to
time series (2
013 )
Homogeneous difference equation
∇4𝑧𝑡 = (1 − 𝐵4)𝑧𝑡 = 𝑧𝑡 − 𝑧𝑡−4
Characteristic equation: 𝑟4 − 1 = 0
Solution is given by 𝑟 = 14
(four roots on the unit
circle)
𝑟1 = 1, 𝑟2 = −1, 𝑟3 = 𝑖, 𝑟4 = −𝑖
Two real roots and two complex conjugates with
modulus 1 and frequency 𝜔 = 𝜋/2.
INTRODUCTION
Therefore,
1. One in the zero frequency (trend)
2. One in the twice-a-year seasonality
3. Associated with once-a-year seasonality
37
Intro
ductio
n to
time series (2
013 )
)1)(1)(1( 24 BBB
2/
38
Intro
ductio
n to
time series (2
013 )
To see this just recall that the complementary function to
the difference equation 𝑧𝑡 = 𝑏𝑧𝑡−1 + 𝑐𝑧𝑡−2 will eventually be
of the form:
𝑧𝑡 = 𝑟𝑡 𝐴 cos 𝜃𝑡 + 𝐵 sin 𝜃𝑡
where 𝑟 is a positive constant and 𝜃 is an angle measured
in radians. 𝐴 and 𝐵 are arbitrary constants to enable the
solution to satisfy any starting point of 𝑧𝑡 .
Hence, when the two solutions of the characteristic
equations are 𝑖 and – 𝑖, sin 𝜃 = 1 and 𝜃 =𝜋
2.
UNIVARIATE ARMA MODELS
“Building block” is the white noise process
Implications:
39
Intro
ductio
n to
time series (2
013 )
),0( 2Niidat
0,...),,/()( 321 ttttt aaaaEaE
0),(),( jttjtt aaCovaaE
221 ,...),/(),( atttjtt aaaVaraaVar
INTRODUCTION
40
Intro
ductio
n to
time series (2
013 )
General ARMA(p,q) processes:
𝜙 𝐵 𝑧𝑡 = 𝑐 + 𝜃(𝐵)𝑎𝑡
𝑧𝑡 is the observable time series.
𝑎𝑡 is sequence of white noise.
𝑐 is the constant term.
INTRODUCTION
Autoregressive polynomial
Moving-average polynomial
The AR and MA polynomials are assumed to have no
common factor
41
Intro
ductio
n to
time series (2
013 )
ppBBBB ...1)( 2
21
qqBBBB ...1)( 2
21
INTRODUCTION
42
Intro
ductio
n to
time series (2
013 )
Stationarity implies that all zeros of 𝜙 𝐵 are
restricted to lie outside the unit circle and in this case:
𝑐 = 1 − 𝜙1 − ⋯− 𝜙𝑝 𝜇
where 𝜇 is the mean of the series.
Overall behavior of the series remains the same
over time.
INTRODUCTION
43
Intro
ductio
n to
time series (2
013 )
Overall behavior of the series remains the same
over time.
In real world, however, time series data often
exhibits a drifting behavior. Nonstationarity can be
modeled allowing some of the zeroes in 𝜙 𝐵 to be
equal to one. Thus,
𝜙 𝐵 (1 − 𝐵𝑑)𝑧𝑡 = 𝑐 + 𝜃(𝐵)𝑎𝑡
This is known as the ARIMA (p,d,q) model.
INTRODUCTION
Some special cases of ARIMA(p,d,q):
AR(1)
MA(1)
ARMA(1,1)
IMA(1,1)
ARIMA(0,1,1)(0,1,1)
44
Intro
ductio
n to
time series (2
013 )
INTRODUCTION
45
Intro
ductio
n to
time series (2
013 )
AR(1) to MA(∞) by recursive substitution.
𝑧𝑡 = 𝜙𝑧𝑡−1 + 𝑎𝑡
𝑧𝑡 = 𝜙 𝜙𝑧𝑡−2 + 𝑎𝑡−1 + 𝑎𝑡 = 𝜙2𝑧𝑡−2 + 𝜙𝑎𝑡−1 + 𝑎𝑡
𝑧𝑡 = 𝜙𝑘𝑧𝑡−𝑘 + 𝜙𝑘−1𝑎𝑡−𝑘+1 + ⋯ + 𝜙2𝑎𝑡−2 + 𝜙𝑎𝑡−1 + 𝑎𝑡
INTRODUCTION
46
Intro
ductio
n to
time series (2
013 )
If 𝜙 < 1, then
𝑧𝑡 = 𝜙𝑗𝑎𝑡−𝑗
∞
𝑗 =0
which is a MA(∞)
INTRODUCTION
Some considerations:
ARMA models are not unique (variety of possible
representations.
Parsimony is always a desirable property.
AR representations are easiest to estimate (OLS) and
to be interpretated.
47
Intro
ductio
n to
time series (2
013 )
PROPERTIES
For simplicity, we will assume zero mean and a starting point m. The ARIMA (p,d,q)
Can be rewritten as:
where,
are r=max(p,q) initial values.
48
Intro
ductio
n to
time series (2
013 )
tt aBzB )()(
aDzD
)',....,( tm zzz )',...,( tm aaa
)'0,...,0,,...,( 1 rmm ww
1,..., rmm ww
PROPERTIES
49
Intro
ductio
n to
time series (2
013 )
1.........
.........
......
.........
...
1
1
1
p
p
D
1.........
.........
......
.........
...
1
1
1
q
q
D
)1)(1( mtmt
)1)(1( mtmt
PROPERTIES
ARMA weights
Where,
50
Intro
ductio
n to
time series (2
013 )
11 DaDDz
1...
...
1
1
11
mt
DD
PROPERTIES
The -weights can be obtained by equating
coefficients of powers of B from the relations:
where
51
Intro
ductio
n to
time series (2
013 )
)()()( BBB
...1)( 221 BBB
PROPERTIES
The ARIMA (p,d,q) model can then be rewritten in the
MA form as:
52
Intro
ductio
n to
time series (2
013 )
mt
hhthtt aaz
1
PROPERTIES
In the same way,
Where,
53
Intro
ductio
n to
time series (2
013 )
aDzDD 11
1...
...
1
1
11
mt
DD
PROPERTIES
With
The AR form of the model is then,
54
Intro
ductio
n to
time series (2
013 )
).()()( BBB
t
mt
hhtht azz
1
PROPERTIES
These two expressions are of fundamental importance in understanding the nature of the model.
The MA form with the weights, shows how the observation is affected by current and past shocks or innovations.
The AR form with the weigths, indicates how the observation is related to its own past values.
55
Intro
ductio
n to
time series (2
013 )
tz
PROPERTIES
Examples : AR(1) model
MA representation:
AR representation:
56
Intro
ductio
n to
time series (2
013 )
....22
11 tttt aaaz
ttt azz 1
PROPERTIES
Examples : MA(1) model
MA representation:
AR representation:
57
Intro
ductio
n to
time series (2
013 ) tjt
jtt azzz .......1
1 ttt aaz
PROPERTIES
Examples : IMA(1,1) model
MA representation:
AR representation:
58
Intro
ductio
n to
time series (2
013 )
ttttt azzzz ...)1()1()1( 32
21
....))(1( 21 tttt aaaz
STATIONARITY CONDITION
Consider,
Since the innovations are assumed normally
distributed, it follows that the observations are also
normally distributed.
It is seen that, if in the characteristic equation:
59
Intro
ductio
n to
time series (2
013 )
mt
hhthtt aaz
1
kpp
kk rArA 0011 ...
STATIONARITY CONDITION
(Where the A’s denote polinomials in k and the r’s are
the distinct zeroes of )
Then as we have that
60
Intro
ductio
n to
time series (2
013 )
)(B
,1jrpoj ,...,1
, mt
,0)( tzE
0
2),cov(h
khhaktt zz
STATIONARITY CONDITION
So that will be stationary in this asymptotic sense.
This is the stationarity condition of an ARMA model,
and is equivalent to require that the zeroes of
are lying outside the unit circle.
61
Intro
ductio
n to
time series (2
013 )
tz
)(B
LAG K AUTOCOVARIANCE
Let us denote,
For an alternative expressions of the autocovariance
function in terms of the ARMA parameters,
62
Intro
ductio
n to
time series (2
013 )
)(),cov(),cov()( kzzzzk kttktt
)...( 11 ptpttkt zzzz
)...( 11 qtqttkt aaaz
LAG K AUTOCOVARIANCE
By taking expectations on both sides and using the
MA form, we obtain, for
Where,
63
Intro
ductio
n to
time series (2
013 )
k
p
hh ghkk
1
)()(
qk
kg
kq
hkhha
k
0
,...2,1,00
2
,0k
AUTOCORRELATION FUNCTION
The autocorrelation function is defined as
By substitution of , we obtain that
64
Intro
ductio
n to
time series (2
013 )
,...2,1,0,)0(
)()( k
kk
)(k
0)0(
)()(1
kg
hkkp
h
kh
AUTOCORRELATION FUNCTION
When , in a MA(q) model, then,
That is, the autocorrelation function cuts off after lag
q.
65
Intro
ductio
n to
time series (2
013 )
0)( B
qk
qkk qq
,0
,)...1()(
1221
PARTIAL AUTOCORRELATION FUNCTION
Intuition
AR(1):
AR(2):
66
Intro
ductio
n to
time series (2
013 )
tttt zzzz 123
tttt zzzz 123
PARTIAL AUTOCORRELATION FUNCTION
The autocorrelation function only takes into account
that and are related in both cases.
But if we want to measure the direct relationship
(without the intermediate ), we found that it is
zero for the AR(1) and different from zero for the
AR(2) model
67
Intro
ductio
n to
time series (2
013 )
tz 2tz
1tz
PARTIAL AUTOCORRELATION FUNCTION
The partial autocorrelation function is, therefore, a
measure of the linear relation among observations k-
periods apart, independently of the intermediate
values.
68
Intro
ductio
n to
time series (2
013 )
PARTIAL AUTOCORRELATION FUNCTION
Consider first an stationary AR(p) model. The
autoregressive coefficients are related to the
autocorrelations by the Yule-Walker equations,
where, is the matrix,
69
Intro
ductio
n to
time series (2
013 )
pppG
pG )( pxp
PARTIAL AUTOCORRELATION FUNCTION
70
Intro
ductio
n to
time series (2
013 )
1)1(...)2()1(
)1(1)2(
)2(1)1(
)1()2(...)1(1
pp
p
p
pp
Gp
PARTIAL AUTOCORRELATION FUNCTION
Regarding this as system of p equations and p
unknowns (the coefficients), the solution is, for p>1,
the ratio of two determinants
71
Intro
ductio
n to
time series (2
013 )
ppp GH /
PARTIAL AUTOCORRELATION FUNCTION
72
Intro
ductio
n to
time series (2
013 )
)()1(...)2()1(
)1(1)2(
)2(1)1(
)1()2(...)1(1
ppp
pp
p
H p
where
PARTIAL AUTOCORRELATION FUNCTION
73
Intro
ductio
n to
time series (2
013 )
•A pxp matrix, and for p=1. This
leads to define, for any stationary model
•Which is known as the Partial Autocorrelation
Function.
)1( p
1/
1)(
kGH
kk
kk
k
PARTIAL AUTOCORRELATION FUNCTION
74
Intro
ductio
n to
time series (2
013 )
It has the property that, for a stationary AR(p)
model,
In other words, vanishes for k>p when the
model is AR(p).
pk
pkk
k0
k
PROPERTIES
75
Intro
ductio
n to
time series (2
013 )
SAF PAF
AR(P) slow decay
towards zero
p different
from zero
MA(q) q different
from zero
slow decay
towards zero
ARMA(p,q) Slow decay
towards zero
Slow decay
towards zero
EXAMPLES
AR(1) model
Model:
Variance:
76
Intro
ductio
n to
time series (2
013 )
ttt azz 1
2222azz
2
22
1
a
z
EXAMPLES
Autocovariance function:
77
Intro
ductio
n to
time series (2
013 )
1)1(
11
0
)(2
2
2
kk
k
k
k a
z
EXAMPLES
Autocorrelation function:
Partial autocorrelation:
78
Intro
ductio
n to
time series (2
013 )
1
1
01
)(
k
k
k
kk
)1()1(
EXAMPLES
AR(2) model
Model:
Variance:
79
Intro
ductio
n to
time series (2
013 )
tttt azzz 2211
21
22
2
2
22
)1(1
1
a
z
EXAMPLES
Autocorrelation function:
80
Intro
ductio
n to
time series (2
013 )
21
11
)(
2
21
2
2
1
k
k
k
22211 kkkk
EXAMPLES
MA(1) model
Model:
Variance:
81
Intro
ductio
n to
time series (2
013 )
1 ttt aaz
2222aaz
)1( 222 az
EXAMPLES
Autocovariance function:
82
Intro
ductio
n to
time series (2
013 )
10
1
0
)( 2
2
k
k
k
k a
z
EXAMPLES
Autocorrelation function:
83
Intro
ductio
n to
time series (2
013 )
10
11
01
)(2
k
k
k
k
SPECIFICATION
The class of ARMA(p,q) models is extensive.
Guidelines are needed in selecting a member of the
class to represent the time series data at hand
84
Intro
ductio
n to
time series (2
013 )
SPECIFICATION
Box and Jenkins (1976) proposed an iterative model
building strategy.
Tentative specification or identification of a model.
Efficient estimation of model parameters.
Diagnostic checking of fitted model for further
improvement.
85
Intro
ductio
n to
time series (2
013 )
SPECIFICATION
Tentative specification. The aim is to employ statistics
that:
1. Can be readily calculated from the
data.
2. Allow the user to tentatively select a
model .
86
Intro
ductio
n to
time series (2
013 )
SPECIFICATION
Three methods:
The Sample Autocorrelation Function
(SAF)
The Sample Partial Autocorrelation
Function (SPAF)
Use of diagnostic tools (AIC,BIC) to
find the best model (TRAMO-
automatic).
87
Intro
ductio
n to
time series (2
013 )
SPECIFICATION
Sample Autocorrelation Function. The SACF of are defined as
with
and the sample mean of the n available observations
88
Intro
ductio
n to
time series (2
013 )
tz
,...2,1/ˆ 0 kCCkk
))((1
zzzzC jt
jn
ttj
z
SPECIFICATION
Properties:
1.For stationary models,
2. When there exists a unit root in the AR polynomial
89
Intro
ductio
n to
time series (2
013 )
nkk ̂
1ˆ p
k
SPECIFICATION
If the SACF of the original series is persistently close
to 1 as k increases, one forms the first difference ,
, and studies its SACF to determine whether further
differencing is called for.
90
Intro
ductio
n to
time series (2
013 )
tz
SPECIFICATION
Once stationary is achieved, a cutting off pattern after,
say a lag “q”, in the SACF will then lead to tentative
specification of a MA(q) model.
91
Intro
ductio
n to
time series (2
013 )
SPECIFICATION
The Sample Partial Autocorrelation Function:
Are obtained by replacing the in
by their sample estimates
92
Intro
ductio
n to
time series (2
013 )
,...2,1ˆ kk
k
1/
1)(
kGH
kk
kk
k
k̂
SPECIFICATION
Properties:
1.For stationary models,
2. The are asymptotically normally distributed
3. For a stationary AR(p) model
93
Intro
ductio
n to
time series (2
013 )
nkk ̂
k̂
pknVar k 1)ˆ(
SPECIFICATION
Properties 1, 2 and 3 make SPACF a convenient tool
for specifying the order or a stationary AR model
(cutting off pattern after lag p)
Not valid for non-stationary models.
94
Intro
ductio
n to
time series (2
013 )
SPECIFICATION
Weakness of the SACF and SPACF.
1. Subjective judgment is often required to decide on
the order of differencing.
2. For stationary ARMA models, both SACF and
SPACF tend to exhibit a gradual “tapering off”
behavior, making the specification very difficult.
95
Intro
ductio
n to
time series (2
013 )
SPECIFICATION
Diagnostic Tools
The program (TRAMO), in an automatic way,
specifies a set of possible models, estimate them and
select the best one based on AIC and BIC criteria.
However an accurate jugment is always necessary to
interprect the output of TRAMO.
96
Intro
ductio
n to
time series (2
013 )