Testing the day of the week effect by GARCH(1,1)

[punk.guyz]

Hi everyone, I'm newbie to Eviews. I'm trying the determine whether the time series data (daily returns of a stock index) is stationary or non-stationary. Afterthat examining the day of the week effect by OLS with dummy variables and GARCH (1,1) model
I attempt to analyse the data by ADF test
I don't under stand how I can choose the lag length in unit root test. I choose the default value of maxlag = 27 and the results are below for the test equation with (1) Intercept, (2) Trend and intercept. Is there anything wrong with these results. As I understand from the ADF test result, the examined time series is stationary. Is there any thing wrong with results?




This is the result of OLS with dummy variables



From this result, I conclude that Tuesday Average return of stock index is statistically significant and negative. While the Friday mean return is statistically significant and positive.



I'm not familiar with the GARCH(1,1). Although I've read several articles about GARCH model and the book named time series analysis using Eviews.
There is only one point I realised that the coefficient of variance equation alpha + beta > 1, this is implausible.

I really need your help. If anyone understand these result please give me some advises. I truly appreciate! Thank you very much.

01-03-2002 ho chi minh vse (pi) - time series data.wf1

This is my workfile

(100.34 KiB) Downloaded 641 times

[trubador]

1) Your results indicate that return series does not appear to have a unit root.
2) Serial correlation among return series is quite strong, so you may need to add several autoregressive lags into your mean equation.
3) Visual inspection indicates that there are breaks in the return series (especially before and after the end-2005).
4) If you have access to EViews version 8, you can search for multiple breaks:

Code: Select all

equation ols.ls r c ols.multibreak(heterr)

5) You can either add these identified break dates as seperate dummies, or simply shorten the sample from the beginning until you obtain a valid GARCH structure.
6) If none of the above helps, then it may indicate that the GARCH structure itself suffers from outliers/breaks or there are more than one GARCH dynamics. And these are even more complicated issues for a beginner...

[punk.guyz]

Thank you for your response Trubador :D
I've tried to remove all the breaks in the data. There was a lot of non-working day on this stock market. However, the result is not better.
Therefore, I decided to shorten the sample by examining the period of 2006-2013. The result become better, indicating that there are a significantly negative Tuesday effect in the market returns. And the sum of alpha + beta < 1.
The result of analysis by GARCH(1,1) and modified-Garch(1,1) (wednesday is not included in the conditional variance equation)



Could you please give me some comments?
Thank you very much!

[trubador]

Results seem allright, but still you should check the usual diagnostics. And note that the constant term in conditional variance equation corresponds to the impact of Wednesday and other coefficents (of Weekdays) simply measure the effect vis-a-vis Wednesday.

[Joshita_22]

Hi there, can someone tell me how to include days of the week in the variance equation to compute Modified GARCH ?

Thank you.