9 box jenkins
TRANSCRIPT
Time series analysisBox-Jenkins method
Chanon Chingchayanurak
Faculty of Business Administration
C.M.U.
Objective
Choose the appropiate model Analyze by using SPSS Diagnosis the models Use model for forecasting
Overview of Box-Jenkins method
Complicate forecasting method, but high accuracy.
Developed by two statisticians, George E.P. Box and Gwilym M. Jenkins.
Published in Time Series Analysis: Forecasting and Control
(1970) The model is called ARMA (p,q) or ARIMA (p,d,q)
Overview of Box-Jenkins method
The Price of Stock A
Date Price (Yt)1 42.35212 43.12563 41.87164 39.41255 39.4589… …
Time series data analysis need >30 samples
Overview of Box-Jenkins method
ARMA (p,q) model
Stationary time series
ARIMA (p,d,q) model
Non-stationary time series
In this session we will focus on ARMA (p,q)
Overview of Box-Jenkins method
Identify Model to Be Tentatively Entertained
Estimate Parameters In Tentatively Entertained
Diagnostic Checking
Use Model for Forecasting
Overview of Box-Jenkins method
Stationary Time Series E(Yt), V(Yt) and Probability distribution
function of Yt is constant at different time period.
Consider
Yt, Yt-1, Yt-2, …, Yt+T-1
and Yt+j, Yt+j-1, Yt+j-2, …, Yt+j+T-1
Overview of Box-Jenkins method
Stationary Time Series Plot the time series data
if its movement has trend Non-stationary
Consider the correlogram of rk
(coefficient autocorrelation at lag K) and K
if rk die down when K is increasing Stationary
ARMA (p,q) Models
p is an order of Autoregressive Models (AR)
q is an order of Moving Average Models (MA)
Parameter in the model equal to p+q+1
(not more than 3)
AR(p) Models
Autoregressive model of order p
where
Yt = dependent variable at time t
= constant
= coefficient (autoregressive parameter)
= error term
tptp2t21t1t Y...YYY
AR(p) Models
AR(1)
or
and
AR(2)
or
and
t1t1t YY tt1 Y)B1( 11
t2t21t1t YYY
t2
21 )BB1( 1,1,1 21221
MA(q) Models
Moving average model of order q
where
Yt = dependent variable at time t
= constant
= coefficient (parameter of moving average)
= error term
qtq2t21t1tt ...Y
MA(q) Models
MA(1)
and
MA(2)
and
1t1ttY 11
1,1,1 21221
2t21t1ttY
ARMA(p,q) Models
Autoregressive and moving average model of
order p and q
ARMA (1,1)
and ,
tptp2t21t1t Y...YYY
qtq2t21t1t ...
11t1t1t YY
11 11
rk and rkk
Consider rk( Coefficient auto correlation) and rkk( Partial coefficient auto correlation)
to identify the appropiate ARMA (p,q) model AR(1) AR(2) MA(1) MA(2) ARMA(1,1)
Coefficient auto correlation: rk
The correlation of the same time series data but one with lagging K period
The variance of rk
n
1t
2t
Kn
1tKtt
k
)YY(
)YY)(YY(r
1k, n
1)r(V k ,...4,3,2k, )r2(1
n
1)r(V
1-k
1j
2
jk
Coefficient auto correlation: rk
Considert 1 2 3 4 5 6 7 8 9 10
Yt 8 5 10 7 12 14 9 15 16 4
Calculate r1,r2,r3
V(r1),V(r2),V(r3)
Coefficient auto correlation: rk
Hypothesis Testing
Test Statistics
Zcalc=
Accept H0 when
0:H
0:H
k1
k0
krn
n
1.96.1r
n
1.96.1 K
Partial coefficient auto correlation: rkk
The correlation of the same time series data but different period (e.g. Lag K) when assume the other different periods remain constant.
,...4,3,2K,rr1
rrrr 1K
1jjj,1K
1K
1jjKj,1Kk
kk
1K, ,rr 1kk
jK,1KKKj,1Kj,K r.rrr n
1)r(V KK and
Partial Coefficient auto correlation: rKK
Hypothesis Testing
Test Statistics
Zcalc=
Accept H0 when
0:H
0:H
KK1
KK0
krn
n
1.96.1r
n
1.96.1 KK
Identify Model with SPSS
Use the SPSS Graphs Time Series Autocorrelation Select the appropiate model from
correlogram.
Overview of Box-Jenkins method
Identify Model to Be Tentatively Entertained
Estimate Parameters In Tentatively Entertained
Diagnostic Checking
Use Model for Forecasting
Estimate Parameter with SPSS
Use the SPSS Analyze Time Series Arima
Or Calculate from the table
Overview of Box-Jenkins method
Identify Model to Be Tentatively Entertained
Estimate Parameters In Tentatively Entertained
Diagnostic Checking
Use Model for Forecasting
Diagnostic Testing
Hypothesis Testing
Test Statistics
Accept H0 when
0:H
0:H
1
0
1n,2
1n,2
ttt
ˆS
ˆt
Overview of Box-Jenkins method
Identify Model to Be Tentatively Entertained
Estimate Parameters In Tentatively Entertained
Diagnostic Checking
Use Model for Forecasting