GARCH with variables

[startz]

Here is simplified example. Still i need to make few corrections in data, but it does not change the point.

Sorry for not being clear. Since we're doing EViews here, I meant an EViews workfile. It would be good to include the equation in the workfile as well as the data. The idea is to make it as low cost a possible for someone to take a look at your problem.

[nerijus]

sorry i put xls, here's eviews file. r are returns and r2 are squared returns. News are dummies [1,0].

[startz]

sorry i put xls, here's eviews file. r are returns and r2 are squared returns. News are dummies [1,0].

This is not an EViews workfile. I assume it's a compressed file of some type.

[nerijus]

I just saved this example in eviews as wf1 file and compressed. How can it be not eviews file. It can be opened from eviews menu file - open - eviews workfile.

[startz]

I just saved this example in eviews as wf1 file and compressed. How can it be not eviews file. It can be opened from eviews menu file - open - eviews workfile.

It's not an EViews workfile because it's compressed.

[nerijus]

Can't eviews read file after extraction? First I rarred file because it was too big to upload it here. Here is smaller example.

[startz]

Can't eviews read file after extraction? First I rarred file because it was too big to upload it here. Here is smaller example.

This makes it much easier to see what's going on.
The difference between the actuals and the residuals depends on the mean equation, not the variance equation. In the subsample you posted the R^2 is almost zero (actually negative). This means you do not explain anything about the level of the return. In particular, note that NEWS is not entered in the mean equation. On the other hand, you do predict quite a bit of variation in the conditional standard deviation.

You have a perfectly reasonable result. It is possible to predict the variance of asset returns but much harder to predict the level.

[nerijus]

R2 are so close to 0 because they are sq. ln returns of 15 min. on EUR/USD. I'm trying to measure impact of news on volatility (somehow quantative, like news increases volatitility 20% etc). As I don't have any background in stats and econometric, it seems somehow too hard for me for now. As I understood from your last post, I should consider building mean equation with news included? Where should I i start when building model witch fits the data best?
Can somebody recommend good book about time series for newbie? :)

[startz]

R2 are so close to 0 because they are sq. ln returns of 15 min. on EUR/USD. I'm trying to measure impact of news on volatility (somehow quantative, like news increases volatitility 20% etc). As I don't have any background in stats and econometric, it seems somehow too hard for me for now. As I understood from your last post, I should consider building mean equation with news included? Where should I i start when building model witch fits the data best?
Can somebody recommend good book about time series for newbie? :)

Perhaps the confusion is that you have two separate equation.



Variable Coefficient Std. Error z-Statistic Prob.

C 5.98E-06 2.70E-06 2.216094 0.0267
AR(2) -0.027492 0.011095 -2.477805 0.0132

Variance Equation

C 2.00E-08 5.84E-10 34.26444 0.0000
RESID(-1)^2 0.227158 0.006963 32.62504 0.0000
GARCH(-1) 0.585539 0.007078 82.73102 0.0000
NEWS 3.51E-07 7.78E-09 45.08955 0.0000

R-squared -0.002031 Mean dependent var 6.11E-07

You are doing a fine job of explaining volatility, and the NEWS variable is significant there. It's the return equation that has no explanatory power. If it's volatility you"re after, the latter is not a problem.

[nerijus]

Ok, why news variable is signifcant here? News coef. should be compared to C constant in variance equation?

[startz]

You might want to take a look at "GARCH 101: The Use of ARCH/GARCH Models in Applied Econometrics," by Rob Engle in the Fall 2001 Journal of Economic Perspectives.

[umaira]

aoa frends! i am not so much familiar with use of E-views. i have to estimate the determinants of Agricultural productivity by using ARCH N GARCH. i have 4 input variables and their affect on 3 crops. i just need guidance to estimate the equation; i mean what will be the format of the equation. and also guide me please whether it is necessary to include exogenous variables in variance equation or not???please help me

[Nguyen Ngoc]

Excuse me, I have a question about this model.
Mean equation: R = α0 + α1*R(t-1) + α2*JAN + α3*ht+ α4*ht*JAN + et ;
Variance equation: ht= b0+b1*h(t-1)+b2*e^2(t-1)
where ht is the variance of et conditional upon the information set at time t -1 and is modeled following an ARMA (1, 1) process.
Can you help me to describe ht*JAN in "mean equation in eviews"? Thank in advance.

[JimForest]

I am performing a wald test on exogenous regressors in the variance equation of a garch model.

EViews output provides t-, f- and chi-squared tests. I am not particularly sure which to report in this scenario.

Anyone familiar with this situation?

[Economist]

GARCH has two parts that you should estimate simultaneously. First part is called "the mean equation", which you can define your stationary time series as univariate and/or as a function of other independent variables. In this part, however, it is assumed that the squared disturbance/error term is not white noise and therefore should be modelled seperately. In the second part, which is called "the variance equation", you can specify the structure of your error term as a GARCH-X process.

Thank you for your explanation, trubador. Nevertheless, I would like to ask you (or anyone who knows about GARCH models) about which variables to add to the mean and variance equations. I research about the natural gas volatility and I have variables, such as oil prices, which affect both the natural gas prices as well as their volatility. So should I add these kinds of variables to the mean equation, or the variance equation, or both? I also have some variables, such as transportation costs, that do not affect natural gas prices and but they might influence their volatility. Where do you think that I should put them? I'd appreciate if you can illuminate me about the variable selection in the mean and variance equations.

Thanks in advance for your time and interest.