Post on 16-Oct-2015
description
Correlogram
Quick Series Statistics- Correlogram.
Put the series name.
Select Level Put number of lags
Press OK
Test the joint significance of autocorrelation up to the given lag-order.
Correlogram of a Non-Stationary /Unit-Root Process
Unit Root Test: Step by Step
Open : Quick Series Statistics-Unit Root Test.
-Enter the variable
Name of the test
1) Lag Selection
First select at Level
First select only intercept
If a series is trend stationary detrend it, or explicitly include trend in your model.
If a series is difference stationary, difference it to make it stationary.
Read t-statistics and p-value;
If p
Granger Causality Test
Quick- Group Statistics-Granger Causality Test
Name Two Series
Select Lag Order
Read F-Statistics and p-value
If p is less than 0.05, null hypothesis is rejected.
VAR Model
Enter at least two variables
Select unrestricted VAR
Select lag order
Lag Length Selection: Go to
View-lag structure-lag length criteria
Select the lag based on the criteria of your choice, where the value of the criteria is minimum.
Granger Causality (Block Exogenity Test)
View-lag structure Granger causality/Block Exogenity Test
Read chi-square and its p-value. Here the null hypothesis of exclusion is rejected, hence US returns cause Indian returns
Here the null hypothesis of exclusion is not rejected; hence Indian returns do not causeUS returns
Impulse Response
View Impulse Response
Define the impulse. Cholesky decomposition is most frequently used.
If Cholesky decomposition is used the sequence of the variable becomes important
Co integration Test
Here select the appropriate model.
If you are not sure you can get summary of the models (Option 6)
In this case use Panetula Principle and select the first model showing cointegration
Results of the first model
If Variables are cointegrated fit a Error correction Model
-Open VAR
-Select Error Correction Model
-Open Cointegration window and select model 1
Ho r=0 rejected
H0 r=1 not rejected
Variables are cointegrated
Results
Cointegration equation normalized to LOGSPOT
Error Correction , since both are significant, there is two way adjustment/causality
Short run VAR,
You may conduct a Granger causality test for this.
Granger Causality Test
Both the hypotheses are rejected, there is two w ay short-run causality
ARCH-GARCH Models
Testing ARCH Effect:
We will estimate AR(1) Model of stock market returns in India (ind) and then we will test the presence of ARCH effect:
Open Quick - Estimate Equation
Get results
AR(1) model
Method of Estimation OLS/ARMA
Go to:
View- Residual Test - ARCH-ML test
Select lag-order (say 1)
Null hypothesis that there is no ARCH effect is strongly rejected, So now estimate ARCH or GARCH model
Estimating ARCH Model
Open-Quick-Estimate Equation
The following dialog box will appear
Enter the variable also the mean equation
Select ARCH from drop-down box
The following dialog Box will appear
If you have to estimate ARCH(1) model make it zero
If you have to estimate ARCH (p) p=1,2,; change this accordingly
Test is there further ARCH Effect in residuals using the procedure discussed earlier. Generally you will observe ARCH effect at higher lag orders as volatility is known to be highly persistent. Therefore it is better to estimate a GARCH model than an ARCH model.
This is the ARCH (1) coefficient
And this is statistically significant.
Estimation of GARCH (1, 1) Model
Results:
ARCH Coefficient
GARCH Coefficient
Test for Residual ARCH effect (selected lag order 10):
If you have to get the estimated conditional volatility graphically; go to
View Conditional SD Graph
You obtain the following graph.
To save the graph in word file go to
Edit - copy
Now ARCH effect is not there in the residuals
To obtain the conditional Volatility as a variable
Go to
Proc Make a GARCH Variance Series
Estimating TRARCH Model
Estimating EGARCH Model:
Change this to 1
Change the model to EARCH using drop down Box
GARCH-X Model
If you want to include some additional variables in volatility equation you can do it by putting the name of the variable in appropriate box. But make sure that the variable should not contain negative values. For example if you have to study the day-of-the week effect in volatility (say Monday effect), make a Monday dummy and put the dummy in the box.
Put additional variable here.