0 Up votes0 Down votes

6 views13 pagesFeb 15, 2012

© Attribution Non-Commercial (BY-NC)

PDF, TXT or read online from Scribd

Attribution Non-Commercial (BY-NC)

6 views

Attribution Non-Commercial (BY-NC)

- Lasso
- Regression
- Looking at Future
- R for Ecological Modelling
- Drivers of NHL Player Salaries
- THE INFLUENCE OF BANK IMAGE ON LOYALTY MODERATED BY SATISFACTION
- Maximum Likelihood
- 2012 Spring BPEA Stock Watson
- Export, Import and Economic Growth
- 321 Project Final
- Hi 2513201329
- Sample Final
- Historical Data in Modelling Warships Battle Damage Survival Probability (Larham)
- final presentation
- regression
- Ch8 Slides
- Applied Psychological Measurement 2015 Wang 119 34
- Price Linkage Between International Price of CPO And Cooking Oil Price in Indonesia
- car prices
- Advance Research Method 1 Midterm Exam Answer Key

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

- LassoUploaded byKelvin Fook
- RegressionUploaded byindra dwinata
- Looking at FutureUploaded byDaniel Barraez
- R for Ecological ModellingUploaded byakshay_paatra
- Drivers of NHL Player SalariesUploaded byJesse Kaufman
- THE INFLUENCE OF BANK IMAGE ON LOYALTY MODERATED BY SATISFACTIONUploaded bydindo_wae
- Maximum LikelihoodUploaded byTewodros Tadesse
- 2012 Spring BPEA Stock WatsonUploaded byalc1b1ad3s
- Export, Import and Economic GrowthUploaded byAlexander Decker
- 321 Project FinalUploaded byAneeta Silvi
- Hi 2513201329Uploaded byAnonymous 7VPPkWS8O
- Sample FinalUploaded byAndrea Momoh
- Historical Data in Modelling Warships Battle Damage Survival Probability (Larham)Uploaded bystugsturmpanzer
- final presentationUploaded byapi-309586551
- regressionUploaded byPhilip Chethalan
- Ch8 SlidesUploaded bykshitij90
- Applied Psychological Measurement 2015 Wang 119 34Uploaded byhadublack
- Price Linkage Between International Price of CPO And Cooking Oil Price in IndonesiaUploaded byFauzul Azhimah
- car pricesUploaded byapi-352507964
- Advance Research Method 1 Midterm Exam Answer KeyUploaded byApril Mergelle Lapuz
- intro712.pdfUploaded bySiri Kalyan
- Regression Analysis of NiftyUploaded byNick
- OlsUploaded byRabia Almamalook
- MIT15_450F10_rec09.pdfUploaded byaluiscg
- _487ba032e00327a71686607be35c016e_Week1_Review_Example_3Uploaded byDuy Khanh Nguyen
- paper_14429985_v30_n8_p892_BellocqUploaded byiris_gomez
- 401 Practice 1Uploaded bykk4jas
- CSSGBUploaded byLekri L
- Cointegration TestsUploaded byEzekiel Fernandez
- PresentationUploaded byJaspreetSingh

- Thesis - Barbara.pdfUploaded byBarbara Wong
- Data Science Course brochureUploaded byRamesh Kummam
- Stepwise Conditional Transformation in Estimation ModeUploaded byAllan Erlikhman
- B11R01 - Descriptive Statistics.docxUploaded byCharles Jebb Belonio Juanitas
- Factor AnalysisUploaded bypopat vishal
- The Impact of Internet Banking of Customer Service(Mutual Trust Bank )Uploaded bySyed Mostakin
- Statistics and Operations Management IUploaded byMike Peña
- MCQ'sUploaded bybitupon.sa108
- Capa training by AEW.pptUploaded byHimanshu sharma
- 2016ASTAT431Uploaded byAndrea Ning
- 37 1497079367_10-06-2017.pdfUploaded byEditor IJRITCC
- Jurnal MetlitUploaded byabiyya salsabil
- For CastingUploaded byManpreet Singh
- Maintenance Model of Hostel Buildings for Effective Performance and AestheticsUploaded byIJMER
- MDM4U OverviewUploaded byAL Rajhi Zakaria
- The Case of Chilli Pepper Value ChainUploaded byGökçe Özaydın
- 6.3 - ANOVA.pptxUploaded byPiyush Singh
- 3. Business - IJBGM - The Development of a Port Performance - Khaled EL.sakty - EgyptUploaded byiaset123
- QTIA ReportUploaded byShakil Aslam
- Testingafter midterm 1.pptxUploaded byNataliAmiranashvili
- ASS1_17_107econometricsUploaded byIves Lee
- Principal Components Factor AnalysisUploaded byIrfan U Shah
- reaction timeUploaded byapi-264015665
- Assignment_Business Research Method (084127-84)Uploaded bySanket Naik
- An Analytical Study on Consumer Buying Behaviourf for Bakery ProductsUploaded byEditor IJTSRD
- Data Science Careers Guide Springboard FinalUploaded byarun.sambargi6220
- Acl ExercisesUploaded byboludu
- Statistical TestsUploaded byAnonymous VKaHkPKs
- Bogdan_Bliken_Qualitative_Research_Educa.pdfUploaded byTheodorus Evan
- 10.1.1.55.7177Uploaded byGordon Lee

## Much more than documents.

Discover everything Scribd has to offer, including books and audiobooks from major publishers.

Cancel anytime.