6 views

Original Title: Arch

Uploaded by Zacharia Ombati Nyabuto

- Optimization of Copper Adsorption by Soil of Polluted Wasteland Using Response Surface Methodology
- Impact of Tourism Consumption on the Social Development of Host Communities in Cross River State, Nigeria
- CE
- SPSS Introduction
- Samples of Output Tables
- Statistical Modelling of Oil Expression from Neem Seed using a Screw Press
- 1-Format.man-An Empirical Analysis Of
- Uni Statistics Project
- L4&5 Multiple Regression 2010B
- Restoring Trust After Fraud Siap Transalte
- Cem
- Whey Protein Nitrogen Index a21a
- 10 Version 1
- Spices Industry
- Dow
- Exports and Imports
- Week8Assessment
- USE OF CORRELATION AND REGRESSION ANALYSIS IN ESTIMATING RELATIVE IMPORTANCE OF FIBER PROPERTIES AFFECTING YARN HAIRINESS IN SOME EGYPTIAN COTTON GENOTYPES.
- 14_Building_Regression_Models_Part1.pdf
- WororomiAMS69!72!2014 Copy

You are on page 1of 13

Nonlinear Univariate

and

Linear Multivariate

Time Series

Seppo Pynn

onen and

Bernd Pape, 2005

University

Statistics, P.O. Box 700, 65100 Vaasa, phone:

(06) 324 8277 email:

bepa@uwasa., URL:

http://www.uwasa./bepa/

1.1 Background

Example. Consider the following daily close-to-close

Nasdaq composite share index values [January 3, 1989

to February 4, 2000]

15

8

10

4

-5

-10

2

1

3.1.1989

2.1.1990

31.12.1990

30.12.1991

28.12.1992

27.12.1993

23.12.1994

22.12.1995

20.12.1996

19.12.1997

21.12.1998

-15

21.12.1999

Day

Return

Log Index

Below are autocorrelations of the log-index. Obviously the persistence of autocorrelations indicate that

the series is integrated. The autocorrelations of the

return series suggest that the returns are stationary

with statistically signicant rst order autocorrelation.

Correlogram of LNSDQ

Correlogram of DNSDQ

Sample: 2276 5080

Included observations: 2805

Autocorrelation

Sample: 2276 5080

Included observations: 2805

Partial Correlation

AC

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

0.998

0.997

0.995

0.993

0.992

0.990

0.988

0.987

0.985

0.983

0.981

0.980

0.978

0.976

0.974

0.973

0.971

0.970

0.968

0.966

0.965

0.963

0.961

0.959

0.958

0.956

0.954

0.953

0.951

0.949

0.948

0.946

0.944

0.943

0.941

0.939

PAC

0.998

-0.010

0.016

0.001

0.013

0.006

-0.018

-0.001

-0.011

0.007

-0.019

0.003

-0.001

-0.003

0.005

0.007

0.016

-0.009

-0.016

0.014

0.018

-0.016

0.001

-0.025

0.005

0.001

-0.002

0.002

-0.003

-0.004

0.000

0.003

0.013

0.000

0.007

0.014

Q-Stat

Prob

2798.4

5587.9

8369.0

11142.

13906.

16663.

19411.

22151.

24882.

27605.

30319.

33025.

35722.

38410.

41090.

43762.

46426.

49081.

51728.

54367.

56998.

59621.

62236.

64842.

67440.

70030.

72611.

75185.

77750.

80306.

82855.

85395.

87927.

90451.

92967.

95476.

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

Autocorrelation

Partial Correlation

AC

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

0.106

0.001

0.008

-0.009

-0.009

-0.012

-0.024

-0.001

-0.006

0.017

-0.004

0.049

0.058

0.009

0.017

-0.028

0.008

0.036

0.040

-0.007

-0.019

-0.046

-0.005

0.027

-0.002

0.013

0.032

0.002

0.020

0.014

0.005

0.004

0.037

-0.032

-0.037

-0.006

PAC

0.106

-0.010

0.009

-0.011

-0.007

-0.010

-0.021

0.004

-0.007

0.019

-0.009

0.051

0.048

-0.002

0.016

-0.032

0.017

0.034

0.038

-0.013

-0.017

-0.044

0.003

0.029

-0.012

0.014

0.025

-0.005

0.021

0.006

-0.002

0.000

0.041

-0.034

-0.022

-0.005

Q-Stat

Prob

31.459

31.465

31.659

31.904

32.123

32.514

34.069

34.071

34.171

35.025

35.079

41.844

51.463

51.686

52.507

54.694

54.875

58.504

63.052

63.174

64.219

70.298

70.379

72.511

72.527

73.018

75.957

75.968

77.127

77.648

77.714

77.767

81.689

84.586

88.467

88.571

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

Figure. Nasdaq Composite index autocorrelations for log levels and log differences (returns)

t

if

E[yt ]

, for all t

cov[yt , yt+k ]

k , for all t

Any series that are not stationary are said to be nonstationary.

Denition Times series yt is said to be integrated of order d,

denoted as yt I(d), if d yt is stationary. Note that if yt is

stationary then yt = 0 yt . Thus for short a stationary series is

denoted as yt I(0), i.e., integrated of order zero.

2

model to the return series

Table. AR(1) estimates.

Dependent Variable: DNSDQ

Method: Least Squares

Sample: 2276 5080

Included observations: 2805

Convergence achieved after 2 iterations

Variable Coecient Std. Error t-Statistic

C

0.086126

0.023048

3.736845

AR(1)

0.105933

0.018782

5.640001

R-squared

Adjusted R-squared

S.E. of regression

Sum squared resid

Log likelihood

Durbin-Watson stat

Inverted AR Roots

0.011221

0.010868

1.091357

3338.542

-4224.341

1.997947

.11

Prob.

0.0002

0.0000

S.D. dependent var

Akaike info criterion

Schwarz criterion

F-statistic

Prob(F-statistic)

0.086119

1.097336

3.013434

3.017668

31.80961

0.000000

Dependent Variable: DNSDQ

Method: Least Squares

Sample: 2276 5080

Included observations: 2805

Convergence achieved after 4 iterations

Variable Coecient Std. Error t-Statistic

C

0.086153

0.022811

3.776796

MA(1)

0.107093

0.018779

5.702685

R-squared

Adjusted R-squared

S.E. of regression

Sum squared resid

Log likelihood

Durbin-Watson stat

Inverted MA Roots

0.011323

0.010970

1.091301

3338.198

-4224.196

2.000051

Prob.

0.0002

0.0000

S.D. dependent var

Akaike info criterion

Schwarz criterion

F-statistic

Prob(F-statistic)

-.11

3

0.086119

1.097336

3.013331

3.017565

32.10153

0.000000

only just marginally better. The residual autocorrelations and related Q-statistics indicate no further autocorrelation left to the series.

Correlogram of Residuals

Sample: 2276 5080

Included observations: 2805

Q-statistic probabilities adjusted for 1 ARMA term(s)

Included observations: 2805

Q-statistic probabilities adjusted for 1 ARMA term(s)

Autocorrelation

Autocorrelation

Partial Correlation

AC

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

0.000

0.001

0.009

-0.010

-0.007

-0.009

-0.023

0.003

-0.008

0.019

-0.011

0.044

0.054

0.001

0.020

-0.031

0.008

0.031

0.038

-0.009

-0.014

-0.045

-0.004

0.029

-0.007

0.011

0.031

-0.003

0.019

0.011

0.004

0.000

0.041

-0.033

-0.034

0.003

PAC

0.000

0.001

0.009

-0.010

-0.007

-0.009

-0.023

0.003

-0.008

0.020

-0.012

0.044

0.053

0.001

0.020

-0.031

0.010

0.032

0.042

-0.008

-0.013

-0.046

-0.005

0.030

-0.010

0.010

0.027

-0.004

0.019

0.008

0.000

-0.004

0.043

-0.027

-0.025

-0.001

Q-Stat

Prob

2.E-05

0.0009

0.2428

0.4990

0.6319

0.8445

2.3176

2.3375

2.5307

3.5945

3.9433

9.5255

17.620

17.624

18.779

21.455

21.634

24.341

28.389

28.622

29.145

34.754

34.794

37.100

37.220

37.533

40.340

40.373

41.442

41.793

41.831

41.831

46.528

49.557

52.794

52.814

0.976

0.886

0.919

0.959

0.974

0.888

0.939

0.960

0.936

0.950

0.574

0.128

0.172

0.174

0.123

0.155

0.110

0.056

0.072

0.085

0.030

0.041

0.032

0.042

0.051

0.036

0.047

0.049

0.059

0.074

0.093

0.047

0.032

0.021

0.027

Partial Correlation

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

0.278

0.272

0.192

0.193

0.217

0.154

0.141

0.145

0.074

0.106

0.107

0.127

0.115

0.124

0.120

0.137

0.133

0.091

0.148

0.076

0.126

0.144

0.105

0.188

0.088

0.120

0.142

0.142

0.120

0.117

0.119

0.106

0.081

0.076

0.087

0.073

PAC

0.278

0.211

0.084

0.089

0.118

0.026

0.022

0.047

-0.039

0.023

0.040

0.051

0.028

0.047

0.032

0.045

0.037

-0.022

0.063

-0.031

0.040

0.065

0.001

0.100

-0.029

0.013

0.046

0.042

-0.014

0.018

0.023

-0.007

-0.013

-0.015

0.007

-0.013

Q-Stat

Prob

216.92

425.28

529.00

633.28

765.20

832.06

888.03

947.14

962.58

994.10

1026.5

1072.0

1109.1

1152.2

1192.7

1245.3

1295.1

1318.5

1380.4

1396.7

1441.8

1500.7

1531.8

1632.3

1654.0

1694.7

1751.8

1808.9

1849.5

1888.4

1928.4

1960.4

1978.9

1995.3

2016.9

2031.9

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

0.000

Figure. Autocorrelations of the squared MA(1) residuals

independence. The autocorrelations of the squared

residuals strongly suggest that there is still left time

dependence in the series. Because squared residuals

are the building blocks of the variance of the series (recall: the sample variance of n observations x1, x2, . . . xn

is s2 = ni=1(xi x

)2/(n 1)), the results suggest that

the variation (volatility) of the series is time dependent.

models.

4

1.2 ARCH-models

The general setup for ARCH models is

yt = xt + ut

with xt = (x1t, x2t, . . . , xpt) , = (1, 2, . . . , p) ,

t = 1, . . . , T , and

ut|Ft1 N (0, ht),

where Ft is the information available at time

t (usually the past values of ut; u1, . . . , ut1),

and

ht = var(ut |Ft1) = + 1u2t1 + 2u2t2 + + q u2tq .

i 0 for all i (to ensure positive variance),

and 1 + +q < 1 (to get stationarity). For

short the model is denoted ut ARCH(q).

The

F. Engle (1982). Autoregressive conditional heteroscedasticity with estimates of the variance of

United Kingdom ination. Econometrica, 50, 987{

1008.

5

Properties of ARCH-processes

Consider (for the sake of simplicity) ARCH(1)

process

ht = + u2

t1

with > 0 and 0 < 1 and ut|ut1

N (0, ht).

(a) ut is white noise

(i) Constant mean (zero):

E[ut] = E[Et1[ut]] = E[0] = 0.

,

=0

expectation given information up to time t

1.

1

2

Et1 [Et2 [ut3 ]] = E [E[ut3 |Ft2 ]|Ft1 ] = E[ut3 |Ft1 ] = Et1 [ut3 ],

6

of iterated expectations, we get

=

2

2

2

u := var[ut] = E[ut ] = E Et1[ut ]

= E[ht] = E[ + u2

t1]

= + E[u2

t1] = + E[ht1]

= + ( + E[u2

t2])

= (1 + ) + 2E[u2

t2]

..

= (1 + + 2 + + n)

]

+ n+1E[u2

, tn11

n

p 0, as

Q

n

i

= limn i=0

.

= 1

(iii) Autocovariances: cov(ut, ut+k ) = 0 for

all k = 0, because using again the law of

iterated expectations we get

= E[utEt+k1(ut+k )) = E(ut 0)

= 0.

7

symmetric, but nonnormal.

(i) Skewness: Exercise, show that E[u3

t ] = 0.

(ii) Kurtosis: Exercise, show that under the

assumption ut|ut1 N (0, ht), and that <

1/3, the kurtosis

2

2

E[u4

.

t] = 3

(1 )2 1 32

2

2 = 3 4,

=

3[var(u

)]

E[u4

t

t] > 3

u

(1 )2

distribution exceed that what it would be,

if ut were normally distributed. Thus the

unconditional distribution of ut is nonnormal

and has fatter tails than a normal distribution

with variance equal to u2 = /(1 ).

8

Consider again the specication of the variance in a general ARCH(q) process:

ht = var(ut |Ft1) = + 1u2t1 + 2u2t2 + + q u2tq .

for the squared residuals, because dening

t = u2

t ht, we can write

2

2

2

u2

t = + 1ut1 + 2ut2 + + q utq + t.

is time dependent (see below), implying that

this is not a stationary process in the sense

dened above. This implies that the conventional estimation procedure in AR-estimation

does not produce optimal results here.

Write

ut

zt =

ht

then zt|Ft1 N(0, 1) implying zt NID(0, 1),

i.e., normally and independently distributed.

Thus we can always write

0

ut = zt ht,

where the zt are independent standard normal

random variables (strict white noise). This

gives us a useful device to check after tting

an ARCH model the adequacy of the specication: Check the autocorrelations of the

squared standardized series.

Returning to the analogy with AR-processes, we note

that using standardized variables we may write

t = u2t ht = zt2ht ht = ht (zt2 1).

independently distributed (iid), that is, in particular

independent of ht , with zero mean and variance 2 (because zt2 is iid 2(1)). So t has zero mean too, but

time-dependent variance, because ht is time-dependent.

10

Given the model

yt = xt + ut

with ut|Ft1 N (0, ht), we have yt|{xt, Ft1}

N (xt, ht), t = 1, . . . , T with conditional pdf

1

f (yt|Ft1) =

2ht

(yt xt )2

2h

t

e

T

with

1

1

1

log ht (yt xt )2/ht,

t () = log(2)

2

2

2

where = ( , , ) .

The maximum likelihood (ML) estimate ^ is the value

maximizing the likelihood function, i.e.,

(^) = max ().

Note: OLS estimates of the regression parameters are

inecient (unreliable) compared to the ML estimates.

11

Consider again for simplicity the ARCH(1) process

with ht = + u2t1. The expected value at time t

of the variance k time steps ahead is

Et (u2t+k ) = Et Et+k1(u2t+k ) = Et (ht+k ) + Et(u2t+k1).

We apply this recursively to get:

Et (u2t+1) = + Et (u2t ) = + u2t = ht+1

Et (u2t+2) = + Et (u2t+1) = + ht+1

Et (u2t+3) = + Et (u2t+2) = + ( + ht+1)

= (1 + ) + 2ht+1

...

Et (u2t+k ) = (1 + + . . . + k2) + k1ht+1

1 k1

+ k1ht+1

=

1

k1

+ k1ht+1

=

1

1 w

W

+ k1 ht+1

=

1

1

k

We note that for long forecast horizons k, the forecasts of the conditional variance ht+k approach the

unconditional variance u2 = /(1 ).

12

- Optimization of Copper Adsorption by Soil of Polluted Wasteland Using Response Surface MethodologyUploaded bySiddhartha Moulik
- Impact of Tourism Consumption on the Social Development of Host Communities in Cross River State, NigeriaUploaded byEditor IJTSRD
- CEUploaded byAndri
- SPSS IntroductionUploaded byniftkislay
- Samples of Output TablesUploaded byFarisha Raz
- Statistical Modelling of Oil Expression from Neem Seed using a Screw PressUploaded byAZOJETE UNIMAID
- 1-Format.man-An Empirical Analysis OfUploaded byImpact Journals
- Uni Statistics ProjectUploaded byNic D
- L4&5 Multiple Regression 2010BUploaded byNguyen Huynh
- Restoring Trust After Fraud Siap TransalteUploaded byarfankaf
- CemUploaded byyusuf arief
- Whey Protein Nitrogen Index a21aUploaded byDinesh Kumar Bansal
- 10 Version 1Uploaded byPETER
- Spices IndustryUploaded bySumitro Dutta
- DowUploaded byAnjali Patel
- Exports and ImportsUploaded byMehul Jain
- Week8AssessmentUploaded byJames Bowman
- USE OF CORRELATION AND REGRESSION ANALYSIS IN ESTIMATING RELATIVE IMPORTANCE OF FIBER PROPERTIES AFFECTING YARN HAIRINESS IN SOME EGYPTIAN COTTON GENOTYPES.Uploaded byIJAR Journal
- 14_Building_Regression_Models_Part1.pdfUploaded byRama Dulce
- WororomiAMS69!72!2014 CopyUploaded byHikmah Fauziah
- WMATA 2005 Development-Related Ridership SurveyUploaded byTerryMaynard
- International Finance and Banking Conference FIBA 2016 XIV-pages-231-239Uploaded bypufu
- Table MatriksUploaded byFajrulFalakh
- Analysis of a Standalone Usage and Limitations .... 2018Uploaded byYoh Sandriano N. Hitang
- Characterizing Spatial Patterns of Precipitation Based on Corrected Trmm 3b43 DataUploaded byJorge Herbozo
- Ensuring the determinant of waqf in Indonesia: does religiosity matter?Uploaded byJEKI
- LAJOP_28_5_1_10_14OTK55UFHUploaded byPhylicia Eu Sheau Yien
- Lecture 14 CUSUM and EWMAUploaded byjinzo88
- Regress GuideUploaded byangelopl
- Financial Aid and CongressUploaded byRyan

- Fallsem2018-19 Eee1007 Eth Tt424 Vl2018191002720 Reference Material i Unit -III BamUploaded byAbhradeep Nag 16BEE0087
- SBE8-PP06Uploaded byAkhila Ayyappan
- FFNN_simonHaykinUploaded byAlexander Jagannathan
- Server NotesUploaded bytofy79
- Short-Term Load Forecasting - Dr. S. N. SinghUploaded byundagunda
- Econometric s Lecture 45Uploaded byKhurram Aziz
- Chapter 5Uploaded byapi-26228448
- 2017 Deutscher - Robust Output Regulation by Observer-based Feedforward Control - IJSSUploaded byiorrus
- Eco No Metrics > Forecasting 2001- Time SeriesUploaded byapi-27174321
- A Guide to Recurrent Neural Networks and BackpropagationUploaded bybottledupthoughtsandfeelings
- CRM Handout - Probability DistributionsUploaded byMönycka Delc
- Copula estimationUploaded byvivipauf
- RJournal_2010-1_Guenther+FritschUploaded byAris Gunaryati
- (Book) Bertsimas, D. & Tsitsiklis, J. N. 19yy Introduction to Linear Optimization - Athena ScientificUploaded byMeysam Shahrbaf
- Time Series Analysis With Arima e28093 Arch013Uploaded byJilani Osmane
- Binomial, Poison and Normal Probability DistributionsUploaded byFrank Venance
- ABM - Lecture 02 - Probability DistributionsUploaded byjerry_jed
- Pm u2 Exercises 5 Nos. Fnl Xlrn Nov 2013Uploaded byHarish Prasad
- Time Series Analysis ARIMA Final RaychellSantosEmbileUploaded byRaychell Manuel Santos-Embile
- Tut 14 Binomial and Poisson Distn_SolutionsUploaded byselmerparis
- Toc(Jelap Book)Uploaded byWaqas Akram Gondal
- Sensitivity Analysis for Neural NetworksUploaded bySERGIO
- Lab Manual Soft ComputingUploaded byAnuragGupta
- STAT2011-2017-exam-formulae.pdfUploaded byeccentricftw4
- UML course pptsUploaded bysatyakiranpriya
- Probability Distributions Wiht SCILABUploaded byMaury Orellana
- Normal+ExponentialUploaded bySarfraz 'Drogba' Shaikh
- IDEF0 and Software Process Engineering ModelUploaded byHouston Mandar Trivedi
- Neural Networks - Comprehensive foundation (Introduction)Uploaded byGuillermo Orozco
- Csci31 ExercisesUploaded byTeodorescu Stefan