Random Walk Without Drift:: Where: Is White Noise Error Term With Mean 0 and Variance

Page 75
Time Series Econometrics
1. Nonstochastic Processes
Random Walk without Drift:

Y1t = Y1t −1 + ut
where: ut is white noise error term with mean = 0 and variance = σ2

Random Walk with Drift:
Y2t = δ + Y2t −1 + ut
EView’s Command
smpl 1 1
genr y1=0
genr y2=0
smpl 1 200
genr u=nrnd
smpl 2 200
series y1=y1(-1)+u
series y2=0.5+y2(-1)+u
smpl 1 200
plot y1 y2
120
100
80
60
40
20
-20
25 50 75 100 125 150 175 200
Y1 Y2
By Tatre Jantarakolica
Page 76
2. Trend Stationary (TS) and Difference Stationary (DS)
Pure random walk β1 = 0 , β2 = 0, β3 = 1

Y1t = Y1t −1 + ut
Random walk with drift β1 ≠ 0 , β2 = 0, β3 = 1
Y2t = β1 + Y2t −1 + ut
Deterministic trend β1 ≠ 0 , β2 ≠ 0, β3 = 0
Y3t = β1 + β 2t + ut
Random walk with drift and deterministic trend β1 ≠ 0 , β2 ≠ 0, β3 = 1
Y4t = β1 + β 2t + Y4t −1 + ut
Deterministic trend with stationary AR(1) component β1 ≠ 0 , β2 ≠ 0, β3 < 1
Y5t = β1 + β 2t + β3Y5t −1 + ut
EView’s Command
smpl 1 1
genr y1=0
genr y2=0
genr y3=0
genr y4=0
genr y5=0
smpl 1 200
genr u=nrnd
genr t=@trend
smpl 2 200
series y1=y1(-1)+u
series y2=0.5+y2(-1)+u
series y3=0.5+0.5*t+u
series y4=0.5+0.5*t+y4(-1)+u
series y5=0.5+0.5*t+0.9*y5(-1)+u
smpl 1 200
genr yhat = 0.5+0.5*t
plot y1 y2 y3 yhat
plot y1 y4 y5
Page 77
120
100
80
60
40
20
-20
25 50 75 100 125 150 175 200
Y1 Y3
Y2 YHAT
12000
10000
8000
6000
4000
2000
-2000
25 50 75 100 125 150 175 200
Y1 Y4 Y5
Page 78
3. Spurious Regression
Yt = Yt −1 + ut
X t = X t −1 + vt
If Xt and Yt are uncorrelated I(1) processes, regression Yt = β1 + β 2 X t + ε t can
lead to spurious problem.
EView’s Command
smpl 1 1
genr y=0
genr x=0
smpl 1 500
genr u=nrnd
genr v=nrnd
smpl 2 500
series y=y(-1)+u
series x=x(-1)+v
smpl 1 500
equation eq_21_7.ls y c x
Dependent Variable: Y
Method: Least Squares
Sample: 1 200
Included observations: 200
Variable Coefficient Std. Error t-Statistic Prob.
C -4.282010 0.418138 -10.24067 0.0000
X 0.492647 0.043581 11.30409 0.0000
R-squared 0.392232 Mean dependent var -0.287086
Adjusted R-squared 0.389163 S.D. dependent var 4.043829
S.E. of regression 3.160498 Akaike info criterion 5.149286
Sum squared resid 1977.772 Schwarz criterion 5.182269
Log likelihood -512.9286 F-statistic 127.7824
Durbin-Watson stat 0.122278 Prob(F-statistic) 0.000000
Page 79
4. Unit Root Test

Example
Autoregressive (Stationary) Series
X1t = 0.05 + 0.95 X1t-1 + u1t
Random Walk with Drift
X2t = 0.05 + X2t-1 + u2t
Explosive Series
X3t = 0.05 + 1.05 X3t-1 + u3t
EView’s Command
smpl 1 1
genr x1 = 0
genr x2 = 0
genr x3 = 0
smpl 1 200
genr u1 = nrnd
genr u2 = nrnd
genr u3 = nrnd
smpl 2 200
series x1 = 0.05+0.95*x1(-1)+u1
series x2 = 0.05+x2(-1)+u2
series x3 = 0.05+1.05*x3(-1)+u3
plot x1
plot x2
plot x3
Graphical Plot
4
-1
-2
-3
25 50 75 100 125 150 175 200
X1
Page 80
35
30
25
20
15
10
-5
25 50 75 100 125 150 175 200
X2
25000
20000
15000
10000
5000
0
25 50 75 100 125 150 175 200
X3
Autocorrelation Function (ACF) and Correlogram

From series Windows (e.g. series X1), go to View and choose Correlogram …
Then, choose number of Lags.
Page 81
Result of Stationary Time Series
Page 82
Result of Nonstationary Time Series
Unit Root Test – ADF Test

From series Windows, go to View and choose Unit Root Test…
Page 83
From Unit Root Test Window, choose Test type (in this case, Augmented Dickey-Fuller),
level of test or Test for unit root in (in this case, Level), equation form or Include in test
equation (in this case, Trend and intercept), and Lag length (in this case, 1) or
suggested by Schwartz Information Criterion by choose Automatic selection.
Unit Root Test of Stationary Series (X1) – Level Test

Null Hypothesis: X1 has a unit root
Exogenous: Constant, Linear Trend
Lag Length: 1 (Fixed)
t-Statistic Prob.*
Augmented Dickey-Fuller test statistic -7.406746 0.0000
Test critical values: 1% level -4.005076
5% level -3.432682
10% level -3.140127
*MacKinnon (1996) one-sided p-values.
Augmented Dickey-Fuller Test Equation

Dependent Variable: D(X1)
Sample(adjusted): 3 200
Included observations: 198 after adjusting endpoints
X1(-1) -0.531530 0.071763 -7.406746 0.0000
D(X1(-1)) 0.060736 0.071894 0.844803 0.3993
C 0.095208 0.141188 0.674332 0.5009
@TREND(2) 0.000561 0.001227 0.456875 0.6483
Page 84
Unit Root Test of Nonstationary Series (X2) – Level Test

Null Hypothesis: X2 has a unit root
Lag Length: 0 (Automatic based on SIC, MAXLAG=14)
t-Statistic Prob.*
5% level -3.432566
10% level -3.140059

Dependent Variable: D(X2)
Sample: 2 200
X2(-1) -0.056878 0.021507 -2.644656 0.0088
C 0.635091 0.213990 2.967855 0.0034
@TREND(2) 0.004893 0.002407 2.032962 0.0434
R-squared 0.035489 Mean dependent var 0.152559
Unit Root Test of I(1) Series (X2) – First Difference Test
Page 85
Null Hypothesis: D(X2) has a unit root

t-Statistic Prob.*
5% level -3.432682
10% level -3.140127

Dependent Variable: D(X2,2)
Sample(adjusted): 3 200
D(X2(-1)) -1.043168 0.071582 -14.57305 0.0000
C 0.215335 0.146020 1.474690 0.1419
@TREND(2) -0.000577 0.001266 -0.455705 0.6491
5. Cointegration Test
Example:
PCEt = β1 + β2PDIt + ut
where PCEt and PDIt are both nonstationary and I(1) processes.
Then, if PCEt and PDIt are cointegrated series, ut = PCEt - β1 - β1PDIt must
be stationary.
Stationarity Test – Unit Root Test Result

Null Hypothesis: PCE has a unit root
Exogenous: Constant
t-Statistic Prob.*
5% level -2.895109
10% level -2.584738
Page 86
Null Hypothesis: D(PCE) has a unit root

Exogenous: Constant
t-Statistic Prob.*
5% level -2.895512
10% level -2.584952
Null Hypothesis: PDI has a unit root

Exogenous: Constant
t-Statistic Prob.*
5% level -2.895109
10% level -2.584738
Null Hypothesis: D(PDI) has a unit root

Exogenous: Constant
t-Statistic Prob.*
5% level -2.895512
10% level -2.584952
Cointegrating Equation
EView’s Command
equation eq1.ls pce c pdi
genr uhat=resid
Dependent Variable: PCE

Sample: 1970:1 1991:4
C -171.4412 22.91725 -7.480880 0.0000
PDI 0.967250 0.008069 119.8712 0.0000
Page 87
Cointegration Test by Testing Stationarity of Residual

Null Hypothesis: UHAT has a unit root
Exogenous: Constant
t-Statistic Prob.*
5% level -2.895109
10% level -2.584738

Dependent Variable: D(UHAT)
Sample(adjusted): 1970:2 1991:4
UHAT(-1) -0.275357 0.073267 -3.758250 0.0003
C -0.441396 2.615711 -0.168748 0.8664
Error Correction Mechanism (ECM)

ΔPCEt = α0 + α1ΔPDIt + α2 ut + εt
EView’s Command
equation eq2.ls d(pce) c d(pdi) uhat(-1)
Dependent Variable: D(PCE)

Sample(adjusted): 1970:2 1991:4
C 11.69183 2.195675 5.324936 0.0000
D(PDI) 0.290602 0.069660 4.171715 0.0001
UHAT(-1) -0.086706 0.054180 -1.600311 0.1133
Page 88
6. Autoregressive Integrated Moving Average (ARIMA) Models
Example
MA(1)
Y1t = 0.1 + u1t+ 0.5 u1t-1
AR(1)
Y2t = 0.1 + 0.5 Y2t-1 + u2t
ARMA(1,1)
Y3t = 0.1 + 0.5 Y3t-1 + u3t + 0.5 u3t-1
ARIMA(1,1,1)
ΔY4t = 0.1 + 0.5 ΔY4t-1 + u4t + 0.5 u4t-1
EView’s Command
smpl 1 1
genr y1 = 0
genr y2 = 0
genr y3 = 0
genr y4 = 0
genr dy4 =0
smpl 1 200
genr u1 = nrnd
genr u2 = nrnd
genr u3 = nrnd
genr u4 = nrnd
smpl 2 200
series y1 = 0.1+ u1 + 0.5*u1(-1)
series y2 = 0.1 + 0.5*y2(-1) + u2
series y3 = 0.1 + 0.5*y3(-1) + u3 + 0.5*u3(-1)
series y4 = 0.1 + y4(-1) + u4 + 0.5*u4(-1)
genr dy4 = y4 – y4(-1)
series y4 = 0.1 + y4(-1) + 0.5*dy4(-1) + u4 + 0.5*u4(-1)
plot y1 y2 y3 y4
Page 89
70
60
50
40
30
20
10
-10
25 50 75 100 125 150 175 200
Y1 Y3
Y2 Y4
Correlogram of MA(1)
Page 90
Correlogram of AR(1)
Page 91
Correlogram of ARMA(1,1)
Page 92
Correlogram of ARIMA(1,1,1)
Page 93
Example: Finding ARIMA process of Price Index (CMI) – Box-Jenkins Methodology

Step 1 & 2:
Identification and Estimation: The first step is to identify the order of
ARIMA(p,d,q). To find the value of p, d, and q, we should firstly identify value of d by
performing unit root test to find which order the series is integrated.
Unit Root Test
Find integrated order of the series using unit root test.
From EViews, Click the series, go to View, then choose unit root (follow the step
of testing unit root). From this example, CMI series is I(1). Thus, d = 1 or the model is
ARIMA(p,1,q).
Page 94
Null Hypothesis: CMI has a unit root

t-Statistic Prob.*
5% level -3.455376
10% level -3.153438
Null Hypothesis: D(CMI) has a unit root

t-Statistic Prob.*
5% level -3.455376
10% level -3.153438
Identify p and q
Next step is to identify order p and q of the model. We should first specify the
maximum lags length to limit the combination of specification of the models to be tested
(for example, in this case, we set maximum lags length = 2). Then, the combinations of
the models to be tested include ARIMA(0,1,1), ARIMA(1,1,0), ARIMA(1,1,1),
ARIMA(1,1,2), ARIMA(2,1,0), ARIMA(2,1,1), and ARIMA(2,1,2). To identify the optimal
order of the lags number of the ARIMA model, Schwarz Information Criteria (SIC) and
Akaike Information Criteria (AIC) can be used by choose the model with the lowest
value of SIC of AIC (in this case, we use SIC as our choosing criteria).
From EViews, click Estimate model, specify the order of ARIMA model. For
example, if the model to be estimated is ARIMA(2,1,2), the command is:
D(CMI) C AR(1) AR(2) MA(1) MA(2)
If the model is ARIMA(1,1,1), the command is:
D(CMI) C AR(1) MA(1)
If the model is ARIMA(1,0,0), the command is:
CMI C AR(1)
For example, if we estimate ARIMA(2,1,2), we can specify the command as
follows:
Page 95
Then, click OK.

Dependent Variable: D(CMI)
Sample (adjusted): 1997M04 2005M06
Included observations: 99 after adjustments
Convergence achieved after 18 iterations
Backcast: 1997M02 1997M03
C 0.360691 0.222549 1.620722 0.1084

AR(1) -0.317276 0.684607 -0.463443 0.6441
AR(2) 0.213323 0.216329 0.986106 0.3266
MA(1) 0.805658 0.687702 1.171521 0.2444
MA(2) 0.024205 0.395562 0.061191 0.9513

Inverted AR Roots .33 -.65

Inverted MA Roots -.03 -.77
Page 96
After estimate all combination of the models, then, compare SIC (or AIC) indices
to identify the optimal lags length (in this case, ARIMA(2,1,1) provide the lowest lags
length).
ARIMA 0,1,1 1,1,0 1,1,1 1,1,2 2,1,0 2,1,1 2,1,2

AIC 3.2428 3.3317 3.2534 3.2432 3.3237 3.2411 3.2836
SIC 3.3204 3.4098 3.3576 3.3735 3.4285 3.3722 3.4409
Then, the selected model is ARIMA(2,1,1)

Dependent Variable: D(CMI)
Sample (adjusted): 1997M04 2005M06
Backcast: 1997M03
C 0.407365 0.239446 1.701278 0.0922

DUM* -5.901357 0.530451 -11.12517 0.0000
AR(1) -0.200610 0.107485 -1.866398 0.0651
AR(2) 0.213423 0.109349 1.951762 0.0539
MA(1) 0.966275 0.022858 42.27297 0.0000

Inverted AR Roots .37 -.57

Inverted MA Roots -.97
*In this example, since there is price war in the market in February, 2002. To eliminate this
effect, we use dummy variable by setting DUM=1 in February, 2002 and DUM=0 in all others.
Page 97
Step 3:
Diagnostic Checking: Since residual terms of ARIMA model are assumed to
be white-noise, we can diagnostically check the estimated model by testing the property
of the estimated residual of the estimated model – using Correlogram.
In this example, correlogram result shows that residual terms of the model are
white-noise process.
Step 4:
Forecasting: The last step is forecasting process. To ensure that the
estimated model provides accurate prediction, forecasting error index, such as, Root
Mean Squares Error (RMSE), or Theil’s Inequality Coefficient, is used. In EViews’
estimated result Windows, from menu bar click Forecast.
Page 98
In Forecast Window, fills in Series to forecast (in this case, CMI), names the
forecasted series (in this case, cmif), choose the forecasting Method (in this case,
Dynamic forecast), and Forecast sample (in this case, 1997m01 2005m05).
180
Forecast: CMIF
Actual: CMI
160
Forecast sample: 1997M01 2005M05
Adjusted sample: 1997M04 2005M05
140 Included observations: 98
120 Root Mean Squared Error 7.026952

Mean Absolute Error 5.278022
Mean Abs. Percent Error 4.837119
100 Theil Inequality Coefficient 0.033602
Bias Proportion 0.413649
80 Variance Proportion 0.026065
Covariance Proportion 0.560286
60
1997 1998 1999 2000 2001 2002 2003 2004
CMIF
Page 99
7. Generalized Autoregressive Conditional Heteroscedastic (GARCH)

Models
Example
SETt = β 0 + β1 IBR + β 2GOLDB + β 3USDS + ε t
where: SETt = Return on Stock Exchange of Thailand (SET).

IBRt = Inter-bank rate.
GOLDBt = Gold price.
USDSt = Exchange rate ($US/฿Baht).
εt = Residual which has GARCH(p,q) process:
σ t2 = α 0 + δ1σ t2−1 + δ 2σ t2− 2 + " + δ pσ t2− p + α1ε t2−1 + α 2ε t2− 2 + " + α qε t2− q
Testing GARCH Effect

Firstly, estimate the model using OLS without GARCH(p,q) process.
EView’s Command
equation modelols.ls set c ibr goldb usds
Then, from EViews’ estimated results Windows, go to Views, choose Residual

Tests, and select ARCH LM Test..
Page 100
Then, in Lag Specification Window, specify the lag number to be tested in Lags to
include: (in this case, Lags to include: = 1).
ARCH Test:
F-statistic 134.4074 Prob. F(1,115) 0.000000

Obs*R-squared 63.05211 Prob. Chi-Square(1) 0.000000
Test Equation:
Dependent Variable: RESID^2
Sample (adjusted): 12/03/2003 5/13/2004
C 180.7126 74.34768 2.430642 0.0166

RESID^2(-1) 0.733750 0.063290 11.59342 0.0000

Sum squared resid 49957771 Schwarz criterion 15.88380
In this case, since p-value of the ARCH effect test (F-statistic or Chi-Square
(Obs*R-squared = (117*0.538907) = 63.05211)) is less than level of significance 0.05,
thus, null hypothesis that there is no ARCH effect is rejected, thus, there exists
significant ARCH effect in this model with 0.05 significant level.
Page 101
Identify Order (p,q) and Estimation

The next step is to identify order of GARCH(p,q) by estimating GARCH models
in several orders and choose the model with the lowest AIC or SIC.
Estimate GARCH(p,q) using MLE
From EViews Equation Estimation Window, go to Estimation settings/Method,
choose ARCH – Autoregressive Conditional Heteroskedasticity.
Page 102
From GARCH Equation Estimation Window, specify order of GARCH(p,q) – p in

GARCH box, and q in ARCH box (in this case, we use GARCH(1,1)).
From estimated results of GARCH models with different orders, choose the model with
the lowest AIC or SIC (in this case, the most appropriated model is GARCH(0,1) or
ARCH(1) model).
GARCH 0,1 1,1 1,2 2,1 2,2
AIC 8.8701 9.0228 8.9792 8.9430 8.9438
SIC 9.0110 9.1872 9.1670 9.1308 9.1551
The estimated result can be shown as follows:
Page 103
Dependent Variable: SET

Method: ML - ARCH (Marquardt) - Normal distribution
Sample (adjusted): 12/02/2003 5/13/2004
Variance backcast: ON
GARCH = C(5) + C(6)*RESID(-1)^2
Coefficient Std. Error z-Statistic Prob.
C 3180.374 198.1774 16.04812 0.0000

IBR -64.52450 41.88617 -1.540473 0.1234
GOLDB 0.058916 0.013876 4.245984 0.0000
USDS -72.51075 2.641941 -27.44602 0.0000
Variance Equation
C 93.71101 38.35904 2.442997 0.0146

RESID(-1)^2 0.933997 0.327187 2.854630 0.0043

Estimate Variance (Volatility) of Dependent Variable from Estimated Model

Variance or volatility of dependent variable (in this case, return on SET) can be
estimated from the estimated result σ2t = 93.711 + 0.934ε2t-1
From EViews estimated result Windows menu bar, go to Proc, choose Make
GARCH Variance Series...
Page 104
Then, in Make GARCH Variance Window, define name of GARCH series in

Conditional Variance: box (in this case, garch01), and click OK.
The estimated variance series is then estimated and keep as garch01.
6000
5000
4000
3000
2000
1000
0
2003M12 2004M01 2004M02 2004M03 2004M04 2004M05
GARCH01

Random Walk Without Drift:: Where: Is White Noise Error Term With Mean 0 and Variance

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Random Walk Without Drift:: Where: Is White Noise Error Term With Mean 0 and Variance

Uploaded by

Copyright:

Available Formats

Page 75

Time Series Econometrics

Random Walk without Drift:

where: ut is white noise error term with mean = 0 and variance = σ2

2. Trend Stationary (TS) and Difference Stationary (DS)

Pure random walk β1 = 0 , β2 = 0, β3 = 1

4. Unit Root Test

Autocorrelation Function (ACF) and Correlogram

Result of Stationary Time Series

Result of Nonstationary Time Series

Unit Root Test – ADF Test

Unit Root Test of Stationary Series (X1) – Level Test

Augmented Dickey-Fuller Test Equation

Unit Root Test of Nonstationary Series (X2) – Level Test

Augmented Dickey-Fuller Test Equation

Unit Root Test of I(1) Series (X2) – First Difference Test

Null Hypothesis: D(X2) has a unit root

Augmented Dickey-Fuller Test Equation

Stationarity Test – Unit Root Test Result

Null Hypothesis: D(PCE) has a unit root

Null Hypothesis: PDI has a unit root

Null Hypothesis: D(PDI) has a unit root

Dependent Variable: PCE

Cointegration Test by Testing Stationarity of Residual

Augmented Dickey-Fuller Test Equation

Error Correction Mechanism (ECM)

Dependent Variable: D(PCE)

6. Autoregressive Integrated Moving Average (ARIMA) Models

Example: Finding ARIMA process of Price Index (CMI) – Box-Jenkins Methodology

Null Hypothesis: CMI has a unit root

Null Hypothesis: D(CMI) has a unit root

Then, click OK.

Variable Coefficient Std. Error t-Statistic Prob.

C 0.360691 0.222549 1.620722 0.1084

R-squared 0.188407 Mean dependent var 0.386105

Inverted AR Roots .33 -.65

ARIMA 0,1,1 1,1,0 1,1,1 1,1,2 2,1,0 2,1,1 2,1,2

Then, the selected model is ARIMA(2,1,1)

Variable Coefficient Std. Error t-Statistic Prob.

C 0.407365 0.239446 1.701278 0.0922

R-squared 0.351049 Mean dependent var 0.386105

Inverted AR Roots .37 -.57

120 Root Mean Squared Error 7.026952

7. Generalized Autoregressive Conditional Heteroscedastic (GARCH)

where: SETt = Return on Stock Exchange of Thailand (SET).

Testing GARCH Effect

Then, from EViews’ estimated results Windows, go to Views, choose Residual

F-statistic 134.4074 Prob. F(1,115) 0.000000

Variable Coefficient Std. Error t-Statistic Prob.

C 180.7126 74.34768 2.430642 0.0166

R-squared 0.538907 Mean dependent var 674.5738

Identify Order (p,q) and Estimation

From GARCH Equation Estimation Window, specify order of GARCH(p,q) – p in

The estimated result can be shown as follows:

Dependent Variable: SET

Coefficient Std. Error z-Statistic Prob.

C 3180.374 198.1774 16.04812 0.0000

C 93.71101 38.35904 2.442997 0.0146

R-squared 0.713516 Mean dependent var 697.0034

Estimate Variance (Volatility) of Dependent Variable from Estimated Model

Then, in Make GARCH Variance Window, define name of GARCH series in

The estimated variance series is then estimated and keep as garch01.