You are on page 1of 24

Time series analysis of

Daily Closing Stock Price Indices of

Standard & Poors 500 (S&P 500)


Course Instructor:
Prof. Samarjit Das

Report by:
Mohar Sen, QE1201

Data
Data analyzed Unadjusted closing price of S&P 500

Vintage 03/01/1950 09/12/2013


Source Yahoo Finance
Number of Observations 16088

Nature Non Seasonal (Daily data)


Variables
Level data series P
First difference Series DP

Methodology

Stationarity
Analysis

Conditional
mean

Raw data plot


Unit root test
Removal of nonstationarity
Confirmation of
stationarity

Box-Jenkins analysis
Estimation
Information criteria
finalizing
specification
Residual diagnostics

Volatility analysis

Conclusion

ARCH test
ARCH/GARCH model
fitting
Information criteria
finalizing
specification
Residual diagnostics

Final specification of
the model
Recommendation, if
any

Plot of Raw Data


P
2,000

Observations
1,600

Uneven trend
Non-Stationary data

1,200

800

400

0
55

60

65

70

75

80

85

90

95

00

05

10

Checking for Trend


Observations

Dependent Variable: P
Method: Least Squares
Date: 12/11/13 Time: 11:32
Sample: 1/03/1950 12/09/2013
Included observations: 16088

Both trend and intercept terms are significant

Variable

Coefficient

Std. Error

t-Statistic

Prob.

C
@TREND

-307.2818
0.092369

3.871625
0.000417

-79.36765
221.5926

0.0000
0.0000

R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
F-statistic
Prob(F-statistic)

0.753242
0.753226
245.5470
9.70E+08
-111367.0
49103.29
0.000000

Mean dependent var


S.D. dependent var
Akaike info criterion
Schwarz criterion
Hannan-Quinn criter.
Durbin-Watson stat

435.6903
494.2939
13.84498
13.84593
13.84529
0.000960

Therefore, we proceed to test for unit root


using ADF test with trend and intercept

ADF test (series p)


Variable
P(-1)
D(P(-1))
D(P(-2))
D(P(-3))
D(P(-4))
D(P(-5))
D(P(-6))
D(P(-7))
D(P(-8))
D(P(-9))
D(P(-10))
D(P(-11))
D(P(-12))
D(P(-13))
D(P(-14))
D(P(-15))
D(P(-16))
D(P(-17))
D(P(-18))
D(P(-19))
D(P(-20))
D(P(-21))
C
@TREND(1/03/1950)
R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
F-statistic
Prob(F-statistic)

Coefficient
-0.000213
-0.065128
-0.042691
-0.017087
-0.009753
-0.038168
-0.012548
-0.030904
0.007349
-0.004196
0.032448
-0.011639
0.048758
0.021209
-0.019073
-0.024191
0.027656
0.018294
-0.047570
0.008477
8.17E-05
-0.046335
-0.145949
4.63E-05
0.020029
0.018624
7.539821
911970.1
-55241.17
14.25548
0.000000

Std. Error
0.000243
0.007889
0.007907
0.007914
0.007906
0.007906
0.007909
0.007907
0.007909
0.007908
0.007898
0.007902
0.007900
0.007909
0.007911
0.007909
0.007911
0.007908
0.007910
0.007918
0.007911
0.007896
0.140840
2.59E-05
Mean dependent var
S.D. dependent var
Akaike info criterion
Schwarz criterion
Hannan-Quinn criter.
Durbin-Watson stat

t-Statistic
-0.874981
-8.255719
-5.399168
-2.159080
-1.233615
-4.827940
-1.586680
-3.908456
0.929223
-0.530685
4.108179
-1.472910
6.172055
2.681615
-2.410950
-3.058820
3.496115
2.313352
-6.014067
1.070654
0.010329
-5.868435
-1.036276
1.790536

Prob.
0.3816
0.0000
0.0000
0.0309
0.2174
0.0000
0.1126
0.0001
0.3528
0.5956
0.0000
0.1408
0.0000
0.0073
0.0159
0.0022
0.0005
0.0207
0.0000
0.2843
0.9918
0.0000
0.3001
0.0734
0.111498
7.611029
6.879767
6.891246
6.883563
1.999061

Null Hypothesis: P has a unit root


Exogenous: Constant, Linear Trend
Lag Length: 21 (Automatic - based on SIC, maxlag=42)

Augmented Dickey-Fuller test statistic


Test critical values:
1% level
5% level
10% level

t-Statistic

Prob.*

-0.874981
-3.958605
-3.410082
-3.126769

0.9572

*MacKinnon (1996) one-sided p-values.

Observations
P-Value>0.05

Null hypothesis rejected: Unit root


present (as expected)
First difference calculated dp=p-p(-1)

Plot of Series dp (after first differencing of p)


DP
120

Observations

80

Apparently stationary
Conditionally
heteroskedastic
Unit Root test required
for confirmation of
stationarity

40

-40

-80

-120
55

60

65

70

75

80

85

90

95

00

05

10

ADF test (series dp)


Variable
DP(-1)
D(DP(-1))
D(DP(-2))
D(DP(-3))
D(DP(-4))
D(DP(-5))
D(DP(-6))
D(DP(-7))
D(DP(-8))
D(DP(-9))
D(DP(-10))
D(DP(-11))
D(DP(-12))
D(DP(-13))
D(DP(-14))
D(DP(-15))
D(DP(-16))
D(DP(-17))
D(DP(-18))
D(DP(-19))
D(DP(-20))
C
@TREND(1/03/1950)

R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
F-statistic
Prob(F-statistic)

Coefficient
-1.208127
0.142840
0.099996
0.082756
0.072855
0.034538
0.021837
-0.009215
-0.002010
-0.006353
0.025942
0.014150
0.062754
0.083808
0.064577
0.040232
0.067737
0.085886
0.038172
0.046510
0.046458
-0.080367
2.67E-05

0.539061
0.538429
7.539766
912013.7
-55241.55
852.8214
0.000000

Std. Error
0.041562
0.040571
0.039534
0.038448
0.037418
0.036323
0.035136
0.033957
0.032778
0.031506
0.030062
0.028576
0.026917
0.025169
0.023271
0.021310
0.019191
0.016976
0.014478
0.011535
0.007894
0.119239
1.29E-05

Mean dependent var


S.D. dependent var
Akaike info criterion
Schwarz criterion
Hannan-Quinn criter.
Durbin-Watson stat

t-Statistic
-29.06799
3.520791
2.529330
2.152420
1.947064
0.950846
0.621521
-0.271356
-0.061307
-0.201653
0.862927
0.495183
2.331407
3.329787
2.774993
1.887948
3.529544
5.059121
2.636591
4.032085
5.884990
-0.674003
2.074122

Prob.
0.0000
0.0004
0.0114
0.0314
0.0515
0.3417
0.5343
0.7861
0.9511
0.8402
0.3882
0.6205
0.0197
0.0009
0.0055
0.0591
0.0004
0.0000
0.0084
0.0001
0.0000
0.5003
0.0381

0.000204
11.09785
6.879690
6.890691
6.883328
1.999071

Null Hypothesis: P has a unit root


Exogenous: Constant, Linear Trend
Lag Length: 20 (Automatic - based on SIC, maxlag=42)

Augmented Dickey-Fuller test statistic


Test critical values:
1% level
5% level
10% level

t-Statistic

Prob.*

-29.06799
-3.958605
-3.410082
-3.126769

0.0000

*MacKinnon (1996) one-sided p-values.

Observations
P-Value<<0.05 => Null hypothesis accepted no unit
root in the series dp
It is established that series dp is stationary. We try to
model it as an ARMA(p,q) process by Box Jenkins
method

Correlogram analysis (series dp)


lags

AC

SQRT(N)*
PAC

PAC

Q-Stat

Prob

lags

AC

SQRT(N)*
PAC

PAC

Q-Stat

Observations

Prob

-0.063

-0.063

-7.99058

64.008

19

0.013

0.01

1.268345

289.88

-0.039

-0.043

-5.45389

87.923

20

0.008

0.003

0.380504

291.04

-0.007

-0.012

-1.52201

88.71

21

-0.048

-0.046

-5.83439

328.79

-0.001

-0.004

-0.50734

88.726

22

0.018

0.01

1.268345

334.3

-0.04

-0.041

-5.20022

114.25

23

0.007

0.004

0.507338

335.19

-0.006

-0.012

-1.52201

114.82

24

0.003

0.004

0.507338

335.36

-0.026

-0.031

-3.93187

125.5

25

-0.01

-0.013

-1.64885

337.03

0.009

0.004

0.507338

126.88

26

-0.009

-0.014

-1.77568

338.43

-0.006

-0.008

-1.01468

127.38

27

0.033

0.035

4.439209

355.86

10

0.034

0.032

4.058705

146.22

28

0.007

0.009

1.141511

356.56

11

-0.02

-0.017

-2.15619

152.52

29

0.019

0.019

2.409856

362.49

12

0.05

0.048

6.088058

192.18

30

0.005

0.013

1.648849

362.84

13

0.019

0.025

3.170863

197.81

31

-0.003

0.006

0.761007

362.99

14

-0.023

-0.017

-2.15619

206.48

32

0.001

-0.002

-0.25367

363

15

-0.029

-0.026

-3.2977

219.8

33

-0.016

-0.014

-1.77568

367.14

16

0.035

0.03

3.805036

239.85

34

-0.066

-0.061

-7.73691

437.2

17

0.014

0.022

2.79036

243.04

35

0.006

-0.004

-0.50734

437.7

18

-0.052

-0.048

-6.08806

287.21

36

0.018

0.009

1.141511

443.17

All ACF values significant


No definite threshold found such
that PACF vanishes above that
threshold
This hints at ARMA(p,q) model and
not AR(p) or MA(q) model

ARMA Estimation Information Criteria


Information
Criteria(SIC)

MA(0)

AR(0)

AIC

6.895772

6.891609

6.890118

SIC

6.896728

6.892565

6.891551

AIC

6.892031

6.889474

6.889569

SIC

6.892986

6.890907

6.891480

AIC

6.890391

6.889632

6.889077

SIC

6.891824

6.891544

6.891466

AR(1)

AR(2)

MA(1)

MA(2)

Observations
ARMA(1,1) appears to be the best
fitted model
The suitability has been in line with
both AIC and SIC
Sufficient data present in the model
so SIC is the best criteria

Estimation of the model


Dependent Variable: DP
Method: Least Squares
Date: 12/11/13 Time: 01:44
Sample (adjusted): 1/05/1950 12/09/2013
Included observations: 16086 after adjustments
Convergence achieved after 8 iterations
MA Backcast: 1/04/1950

Observations
As expected, all the terms are significant in
the ARMA(1,1) model

Variable

Coefficient

Std. Error

t-Statistic

Prob.

C
AR(1)
MA(1)

0.111277
0.607851
-0.672801

0.049883
0.056980
0.053089

2.230768
10.66776
-12.67311

0.0257
0.0000
0.0000

R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
F-statistic
Prob(F-statistic)
Inverted AR Roots
Inverted MA Roots

0.006645
0.006522
7.581454
924425.8
-55409.04
53.79525
0.000000
.61
.67

Mean dependent var


S.D. dependent var
Akaike info criterion
Schwarz criterion
Hannan-Quinn criter.
Durbin-Watson stat

0.111371
7.606298
6.889474
6.890907
6.889948
2.004706

The Durbin Watson Statistic is 2. This


suggests that the autocorrelation among
residuals may be small
This is in fact, confirmed by the correlogram
Q-statistics of the residuals

Correlogram Q statistic (of residuals)


lags

AC

SQRT(N)*
PAC

PAC

Q-Stat

Prob

lags

AC

SQRT(N)*
PAC

PAC

Q-Stat

Observations

Prob

-0.002

-0.002

-0.25366

0.0897

19

0.01

0.011

1.395137

183.19

-0.001

-0.001

-0.12683

0.1221

20

0.006

0.001

0.126831

183.72

0.015

0.015

1.902459

3.5484

0.06

21

-0.047

-0.045

-5.70738

219.54

0.011

0.011

1.395137

5.5919

0.061

22

0.016

0.014

1.775628

223.78

-0.031

-0.031

-3.93175

20.833

23

0.007

0.006

0.760984

224.56

-0.002

-0.003

-0.38049

20.928

24

0.003

0.004

0.507322

224.74

-0.022

-0.022

-2.79027

28.685

25

-0.009

-0.011

-1.39514

226

0.011

0.011

1.395137

30.563

26

-0.007

-0.01

-1.26831

226.81

-0.002

-0.001

-0.12683

30.637

27

0.033

0.037

4.692732

244.66

10

0.036

0.035

4.439071

50.953

28

0.009

0.009

1.141475

245.98

11

-0.014

-0.014

-1.77563

54.261

29

0.02

0.017

2.15612

252.52

12

0.05

0.049

6.214699

95.031

30

0.005

0.009

1.141475

252.99

13

0.02

0.02

2.536612

101.69

31

-0.003

0.002

0.253661

253.18

14

-0.021

-0.021

-2.66344

108.6

32

-0.002

-0.006

-0.76098

253.22

15

-0.027

-0.025

-3.17076

120.08

33

-0.019

-0.018

-2.28295

259.02

16

0.033

0.03

3.804918

137.63

34

-0.066

-0.06

-7.60984

328.79

17

0.012

0.017

2.15612

140.07

35

0.003

0.001

0.126831

328.89

18

-0.051

-0.05

-6.34153

181.63

36

0.017

0.013

1.648798

333.34

All ACF values from 5th lag onwards


significant
No definite threshold found such
that PACF vanishes above that
threshold

Correlogram Q statistic (of squared residuals)


lags

AC

SQRT(N)*
PAC

PAC

Q-Stat

Prob

lags

AC

SQRT(N)*
PAC

PAC

Q-Stat

Observations

Prob

0.236

0.236

29.93202

894.26

19

0.209

-0.026

-3.2976

24280

0.387

0.35

44.39071

3298.4

20

0.253

0.03

3.804918

25307

0.243

0.123

15.60016

4251.7

21

0.289

0.105

13.31721

26653

0.297

0.139

17.62945

5671.5

22

0.223

-0.013

-1.6488

27456

0.323

0.189

23.97098

7346.1

23

0.242

-0.004

-0.50732

28402

0.311

0.137

17.37579

8899.2

24

0.191

-0.02

-2.53661

28991

0.303

0.098

12.4294

10377

25

0.209

-0.012

-1.52197

29696

0.295

0.089

11.28792

11777

26

0.205

-0.013

-1.6488

30372

0.283

0.067

8.49765

13068

27

0.276

0.079

10.01962

31600

10

0.311

0.094

11.92208

14621

28

0.233

0.023

2.917104

32474

11

0.307

0.089

11.28792

16138

29

0.214

-0.02

-2.53661

33215

12

0.287

0.045

5.707377

17464

30

0.185

-0.022

-2.79027

33768

13

0.235

-0.027

-3.42443

18356

31

0.194

-0.006

-0.76098

34373

14

0.22

-0.039

-4.94639

19138

32

0.251

0.067

8.49765

35388

15

0.237

0.001

0.126831

20041

33

0.2

-0.005

-0.63415

36032

16

0.271

0.051

6.46836

21225

34

0.232

0.03

3.804918

36902

17

0.269

0.045

5.707377

22387

35

0.169

-0.018

-2.28295

37360

18

0.272

0.047

5.961038

23577

36

0.208

0.013

1.648798

38055

All ACF values significant

No definite threshold found such


that PACF vanishes above that
threshold
However, square of the lag values
are larger (in ACF)
This suggests ARCH type modelling
is more appropriate

ARCH Heteroskedasticity test


Heteroskedasticity Test: ARCH
F-statistic
Obs*R-squared
Variable

C
RESID^2(-1)
RESID^2(-2)
RESID^2(-3)
RESID^2(-4)
RESID^2(-5)
RESID^2(-6)
RESID^2(-7)
RESID^2(-8)
RESID^2(-9)

R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
F-statistic
Prob(F-statistic)

Observations
624.6957
4167.458
Coefficient

11.60923
0.013088
0.213112
0.016465
0.062821
0.147233
0.111116
0.080432
0.087330
0.066907

0.259219
0.258804
246.3978
9.75E+08
-111342.5
624.6957
0.000000

Prob. F(9,16067)
Prob. Chi-Square(9)
Std. Error

2.045328
0.007872
0.007842
0.007996
0.007948
0.007879
0.007948
0.007996
0.007842
0.007872

Mean dependent var


S.D. dependent var
Akaike info criterion
Schwarz criterion
Hannan-Quinn criter.
Durbin-Watson stat

t-Statistic

0.0000
0.0000
Prob.

5.675975
1.662714
27.17408
2.059229
7.903630
18.68761
13.97978
10.05965
11.13558
8.499328

0.0000
0.0964
0.0000
0.0395
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000

57.49988
286.2004
13.85239
13.85717
13.85397
2.012623

We reaffirm the ARCH effect using the


Heteroskedasticity ARCH test
Clearly, all residue lags seem significant (upto
lag 9)
This hints that we may have to opt for a
GARCH(p,q) model
Hence, we have not tested for pure ARCH
models any further

GARCH(1,1)
Dependent Variable: DP
Method: ML - ARCH (Marquardt) - Normal distribution
Date: 12/11/13 Time: 03:05
Sample (adjusted): 1/05/1950 12/09/2013
Included observations: 16086 after adjustments
Convergence achieved after 32 iterations
MA Backcast: 1/04/1950
Presample variance: backcast (parameter = 0.7)
GARCH = C(4) + C(5)*RESID(-1)^2 + C(6)*GARCH(-1)
Variable
C
AR(1)
MA(1)

P-Value of Coefficient of GARCH(-1) is more


less than 0.05
Thus we conclude volatility is of GARCH kind.

Coefficient
Std. Error
0.018850
0.003353
-0.141277
0.075434
0.242258
0.073980
Variance Equation

C
RESID(-1)^2
GARCH(-1)

0.000213
0.066426
0.939441

R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
Durbin-Watson stat
Inverted AR Roots
Inverted MA Roots

Observations

-0.022203
-0.022330
7.690752
951271.9
-30561.24
2.322505
-.14
-.24

3.29E-05
0.001506
0.001222

Mean dependent var


S.D. dependent var
Akaike info criterion
Schwarz criterion
Hannan-Quinn criter.

z-Statistic
5.622112
-1.872843
3.274633

Prob.
0.0000
0.0611
0.0011

6.465720
44.10476
768.9159

0.0000
0.0000
0.0000

0.111371
7.606298
3.800477
3.803344
3.801425

Thus we do not check for ARCH. We check for


the best form of GARCH that fits the data.

GARCH(1,2)
Dependent Variable: DP
Method: ML - ARCH (Marquardt) - Normal distribution
Date: 12/11/13 Time: 03:12
Sample (adjusted): 1/05/1950 12/09/2013
Included observations: 16086 after adjustments
Convergence achieved after 33 iterations
MA Backcast: 1/04/1950
Presample variance: backcast (parameter = 0.7)
GARCH = C(4) + C(5)*RESID(-1)^2 + C(6)*RESID(-2)^2 + C(7)*GARCH(-1)
Variable

Coefficient

C
AR(1)
MA(1)

C
RESID(-1)^2
RESID(-2)^2
GARCH(-1)
R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
Durbin-Watson stat
Inverted AR Roots
Inverted MA Roots

the coefficients of RESID(-1)^2, GARCH(-1) and


GARCH(-2) are all significant at 5% level of
significance.

Std. Error

z-Statistic

Prob.

0.019100
0.003404
-0.137936
0.072289
0.242221
0.070576
Variance Equation

5.610811
-1.908117
3.432082

0.0000
0.0564
0.0006

0.000161
0.107198
-0.050822
0.948322

6.145381
24.13017
-9.928459
621.2227

0.0000
0.0000
0.0000
0.0000

-0.023299
-0.023427
7.694877
952292.6
-30545.67
2.328892
-.14
-.24

Observations

2.62E-05
0.004442
0.005119
0.001527
Mean dependent var
S.D. dependent var
Akaike info criterion
Schwarz criterion
Hannan-Quinn criter.

0.111371
7.606298
3.798666
3.802010
3.799772

GARCH(2,2)
Dependent Variable: DP
Method: ML - ARCH (Marquardt) - Normal distribution
Date: 12/11/13 Time: 03:19
Sample (adjusted): 1/05/1950 12/09/2013
Included observations: 16086 after adjustments
Convergence achieved after 36 iterations
MA Backcast: 1/04/1950
Presample variance: backcast (parameter = 0.7)
GARCH = C(4) + C(5)*RESID(-1)^2 + C(6)*RESID(-2)^2 + C(7)*GARCH(-1) + C(8)*GARCH(-2)

Observations
the coefficients of RESID(-1)^2, RESID(-2)^2,
GARCH(-1) and GARCH(-2) are all significant at
5% level of significance.

Variable

Coefficient

Std. Error

z-Statistic

Prob.

C
AR(1)
MA(1)

0.020273
-0.137986
0.242984

0.003412
0.072567
0.070989

5.940926
-1.901495
3.422855

0.0000
0.0572
0.0006

8.60E-07
0.002502
0.002442
0.006706
0.006578

5.000378
34.00018
-34.23010
279.0638
-132.6706

0.0000
0.0000
0.0000
0.0000
0.0000

Variance Equation
C
RESID(-1)^2
RESID(-2)^2
GARCH(-1)
GARCH(-2)
R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
Durbin-Watson stat
Inverted AR Roots
Inverted MA Roots

4.30E-06
0.085055
-0.083594
1.871332
-0.872703
-0.023536
-0.023663
7.695767
952512.9
-30512.79
2.330303
-.14
-.24

Mean dependent var


S.D. dependent var
Akaike info criterion
Schwarz criterion
Hannan-Quinn criter.

0.111371
7.606298
3.794703
3.798525
3.795967

we cannot draw any inference talking about


the best fitted volatility model to the data
To see which of these models best fit the data,
we consider the minimum AIC and BIC values.

GARCH(2,3)
Dependent Variable: DP
Method: ML - ARCH (Marquardt) - Normal distribution
Date: 12/11/13 Time: 12:23
Sample (adjusted): 1/05/1950 12/09/2013
Included observations: 16086 after adjustments
Convergence achieved after 179 iterations
MA Backcast: 1/04/1950
Presample variance: backcast (parameter = 0.7)

Observations
the coefficients of RESID(-1)^2, RESID(-2)^2,
RESID(-3)^2, GARCH(-1) and GARCH(-2) are all
significant at 5% level of significance.

GARCH = C(4) + C(5)*RESID(-1)^2 + C(6)*RESID(-2)^2 + C(7)*RESID(-3)^2+ + C(8)*GARCH(-1) + C(9)*GARCH(-2)


Variable

Coefficient

C
AR(1)
MA(1)
C
RESID(-1)^2
RESID(-2)^2
RESID(-3)^2
GARCH(-1)
GARCH(-2)
R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
Durbin-Watson stat
Inverted AR Roots
Inverted MA Roots

-.14
-.25

Std. Error

z-Statistic

Prob.

0.019824
0.003325
-0.142799
0.071582
0.248253
0.069856
Variance Equation
1.95E-06
4.27E-07
0.103399
0.004377
-0.133792
0.009751
0.031082
0.005770
1.904230
0.005893
-0.904878
0.005823
-0.023694 Mean dependent var
-0.023822 S.D. dependent var
7.696361 Akaike info criterion
952660.1 Schwarz criterion
-30508.72 Hannan-Quinn criter.
2.331385

5.962788
-1.994902
3.553797

0.0000
0.0461
0.0004

4.569525
23.62315
-13.72090
5.386980
323.1577
-155.4076

0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.111371
7.606298
3.794321
3.798621
3.795743

we cannot draw any inference talking about


the best fitted volatility model to the data
Likewise, all models upto GARCH(3,3) have
been estimated, though, they have not been
pasted in this presentation for brevity.
To see which of these models best fit the data,
we consider the minimum AIC and BIC values.

Volatility Estimation Information Criteria


Information
Criteria(SIC)

AIC

BIC

GARCH(1,1)

3.800477

3.803344

GARCH (2,2) and GARCH (2,3) appear


to be the best fitted models

GARCH(1,2)

3.798666

3.802010

We proceed to estimate both models

GARCH(1,3)

3.798186

3.802009

The final model selection will depend


upon the residual diagnostics

GARCH(2,1)

3.800477

3.802872

GARCH(2,2)

3.794703

3.798525

GARCH(2,3)

3.794321

3.798621

GARCH (3,1)

3.798906

3.802729

GARCH(3,2)

3.799317

3.803617

GARCH(3,3)

3.799790

3.804568

Observations

GARCH (2,3)
Estimation
Variable
C
AR(1)
MA(1)

C
RESID(-1)^2
RESID(-2)^2
RESID(-3)^2
GARCH(-1)
GARCH(-2)

Correlogram Q statistics of Residuals


Coefficient
0.019824
-0.142799
0.248253

Std. Error
0.003325
0.071582
0.069856

Variance Equation
1.95E-06
4.27E-07
0.103399
0.004377
-0.133792
0.009751
0.031082
0.005770
1.904230
0.005893
-0.904878
0.005823

z-Statistic
5.962788
-1.994902
3.553797

4.569525
23.62315
-13.72090
5.386980
323.1577
-155.4076

Prob.
0.0000
0.0461
0.0004

0.0000
0.0000
0.0000
0.0000
0.0000
0.0000

R-squared

-0.023694

Mean dependent var

0.111371

Adjusted R-squared
S.E. of regression
Sum squared resid

-0.023822
7.696361
952660.1

S.D. dependent var


Akaike info criterion
Schwarz criterion

7.606298
3.794321
3.798621

Log likelihood
Durbin-Watson stat

-30508.72
2.331385

Hannan-Quinn criter.

3.795743

lags

AC

SQRT(N)*
PAC
Q-Stat

PAC

lags

Prob

AC

SQRT(N)*
PAC
Q-Stat

PAC

Prob

0.01

0.01 1.268306

1.5477

19

0.003

0.003 0.380492

25.131

0.092

0.002

0.002 0.253661

1.5945

20

0.005

0.004 0.507322

25.486

0.112

-0.006

-0.006 -0.76098

26.126

0.127

0.015

1.5952

0.207

21

0.015 1.902459

5.2146

0.074

22

-0.01

-0.01 -1.26831

27.786

0.115

-0.003

-0.004 -0.50732

27.97

0.141

0.012 1.521967

30.601

0.105

-0.004

-0.004 -0.50732

5.4646

0.141

23

-0.014

-0.014 -1.77563

8.698

0.069

24

0.013

-0.003

-0.002 -0.25366

8.8176

0.117

25

-0.013

-0.013

-1.6488

33.45

0.074

0.008

0.008 1.014645

9.9152

0.128

26

-0.016

-0.015 -1.90246

37.473

0.039

0.006

0.007 0.887814

38.039

0.046

-0.003

-0.003 -0.38049

10.031

0.187

27

10

0.018

0.019 2.409781

15.317

0.053

28

0.005

0.004 0.507322

38.424

0.055

11

-0.007

-0.008 -1.01464

16.206

0.063

29

0.008

0.009 1.141475

39.55

0.056

12

0.013

0.013 1.648798

19.079

0.039

30

0.004

13

0.005

0.005 0.634153

19.468

0.053

31

-0.013

14

-0.004

-0.005 -0.63415

15

16

0.005 0.634153
-0.013

39.77

0.069

-1.6488

42.344

0.052

19.747

0.072

32

0.005

0.004 0.507322

42.673

0.063

19.748

0.102

33

0.011

0.011 1.395137

44.582

0.054

0.011

0.011 1.395137

21.735

0.084

34

-0.012

-0.012 -1.52197

46.866

0.044

17

-0.01

-0.011 -1.39514

23.501

0.074

35

-0.007

-0.006 -0.76098

47.736

0.047

18

-0.01

-0.009 -1.14148

25.015

0.07

36

0.001

0.001 0.126831

47.77

0.059

GARCH (2,3)
Estimation
Variable
C
AR(1)
MA(1)

Correlogram Q statistics of Residuals


Coefficient

Std. Error

z-Statistic

Prob.

0.020273
-0.137986
0.242984

0.003412
0.072567
0.070989

5.940926
-1.901495
3.422855

0.0000
0.0572
0.0006

Variance Equation
C
RESID(-1)^2
RESID(-2)^2
GARCH(-1)
GARCH(-2)
R-squared
Adjusted R-squared
S.E. of regression
Sum squared resid
Log likelihood
Durbin-Watson stat

4.30E-06
0.085055
-0.083594
1.871332
-0.872703
-0.023536
-0.023663
7.695767
952512.9
-30512.79
2.330303

8.60E-07
0.002502
0.002442
0.006706
0.006578

5.000378
34.00018
-34.23010
279.0638
-132.6706

Mean dependent var


S.D. dependent var
Akaike info criterion
Schwarz criterion
Hannan-Quinn criter.

0.0000
0.0000
0.0000
0.0000
0.0000
0.111371
7.606298
3.794703
3.798525
3.795967

lags

AC

SQRT(N)*
PAC
Q-Stat

PAC

Lags

Prob

AC

0.01

0.01 1.268306

1.5963

19

0.001

0.001 0.126831

1.6214

20

1.6214

0.203

21

SQRT(N)*
PAC
Q-Stat

PAC
0.002

Prob

0.003 0.380492

24.605

0.104

0.005

0.004 0.507322

24.978

0.126

-0.006

-0.006 -0.76098

25.604

0.142

0.014

0.014 1.775628

4.9305

0.085

22

-0.01

-0.01 -1.26831

27.214

0.129

-0.004

-0.004 -0.50732

5.1411

0.162

23

-0.003

-0.004 -0.50732

27.395

0.158

-0.014

-0.014 -1.77563

8.5074

0.075

24

0.013

0.013 1.648798

30.322

0.111

-0.003

-0.003 -0.38049

8.6544

0.124

25

-0.014

-0.014 -1.77563

33.438

0.074

0.008

0.008 1.014645

9.6626

0.14

26

-0.016

-0.016 -2.02929

37.617

0.038

-0.002

-0.002 -0.25366

9.7411

0.204

27

0.006

0.007 0.887814

38.194

0.044

10

0.018

0.018 2.282951

14.693

0.065

28

0.005

0.004 0.507322

38.651

0.053

11

-0.007

-0.008 -1.01464

15.589

0.076

29

0.008

0.008 1.014645

39.671

0.055

12

0.013

0.013 1.648798

18.335

0.05

30

0.004

0.005 0.634153

39.933

0.067

13

0.005

0.005 0.634153

18.735

0.066

31

-0.013

-1.6488

42.486

0.051

14

-0.004

-0.005 -0.63415

19.059

0.087

32

0.005

0.004 0.507322

42.841

0.061

15

19.059

0.121

33

0.011

0.011 1.395137

44.682

0.053

16

0.011

0.011 1.395137

21.107

0.099

34

-0.012

-0.012 -1.52197

46.952

0.043

17

-0.011

-0.011 -1.39514

22.955

0.085

35

-0.007

-0.006 -0.76098

47.832

0.046

18

-0.01

-0.009 -1.14148

24.507

0.079

36

0.001

0.001 0.126831

47.86

0.058

-0.013

The best fit model


lags

AC

SQRT(N)*
PAC

PAC

Q-Stat

Prob

lags

AC

SQRT(N)*
PAC

PAC

Q-Stat

Observations

Prob

0.011

0.011

1.395137

2.0998

19

0.002

0.002

0.253661

11.368

0.837

0.002

0.002

0.253661

2.1909

20

-0.002

-0.002

-0.25366

11.424

0.876

0.001

0.001

0.126831

2.2065

0.137

21

-0.002

-0.001

-0.12683

11.46

0.907

-0.002

-0.002

-0.25366

2.2434

0.326

22

-0.005

-0.005

-0.63415

11.849

0.921

2.2444

0.523

23

11.849

0.944

-0.01

-0.01

-1.26831

3.7929

0.435

24

0.002

0.001

0.126831

11.894

0.96

-0.012

-0.012

-1.52197

6.194

0.288

25

-0.001

-0.001

-0.12683

11.921

0.972

-0.001

-0.001

-0.12683

6.2132

0.4

26

-0.022

-0.021

-2.66344

19.371

0.732

0.01

0.01

1.268306

7.7643

0.354

27

-0.005

-0.004

-0.50732

19.765

0.759

10

0.001

7.769

0.456

28

-0.008

-0.008

-1.01464

20.862

0.749

11

-0.002

-0.002

-0.25366

7.816

0.553

29

0.001

0.001

0.126831

20.871

0.792

12

-0.006

-0.006

-0.76098

8.3879

0.591

30

-0.008

-0.008

-1.01464

21.971

0.783

13

-0.004

-0.004

-0.50732

8.619

0.657

31

-0.005

-0.005

-0.63415

22.344

0.806

14

0.002

0.002

0.253661

8.6817

0.73

32

-0.007

-0.008

-1.01464

23.234

0.806

15

-0.005

-0.005

-0.63415

9.1452

0.762

33

0.009

0.009

1.141475

24.675

0.782

16

-0.004

-0.004

-0.50732

9.4626

0.8

34

0.004

0.003

0.380492

24.903

0.81

17

-0.009

-0.009

-1.14148

10.793

0.767

35

0.008

0.008

1.014645

25.907

0.805

18

-0.005

-0.005

-0.63415

11.272

0.792

36

0.002

0.002

0.253661

25.965

0.837

The Residual diagnostics (previous


slides) hint that GARCH(2,3) is a better
fit compared to GARCH(2,2)
We check the squared residuals of
GARCH(2,3) to affirm the goodness of fit
The correlogram affirms our result that
GARCH (2,3) is indeed the best fit model

Conclusion
The data has significant stochastic trend which got removed after first
differencing.
No presence of seasonality in the data.
The best fitted conditional mean model is ARMA(1,1).

Heteroskedasticity exists in the data.


The best fitted volatility model is GARCH(2,3).

DATA SOURCE:
Yahoo Finance
SOFTWARE USED:
Eviews 7
Microsoft Office 2013

Reference:
ARCH/GARCH Models by Prof. Samarjit Das

You might also like