I've wondered why GARCH estimation in EViews does not automatically ignore missing values the way most other procedures do? Eg, if I set a smpl statement to ignore NA values why can't the GARCH routine then treat the filtered series as a continuous one?
Thanks.
NAs in GARCH
Moderators: EViews Gareth, EViews Moderator

 Fe ddaethom, fe welon, fe amcangyfrifon
 Posts: 11592
 Joined: Tue Sep 16, 2008 5:38 pm
Re: NAs in GARCH
It isn't so much that GARCH requires no NAs, but rather GARCH requires a continuous sample (which is a stronger assumption than no NAs!).
This is simply because the GARCH estimation process relies heavily on the fact that each observation has adjacent observations available. GARCH is a dynamic process, each observation depends upon the immediate previous observations.
Most estimators in EViews that follow a dynamic process like GARCH require continuous samples. For example, you cannot estimate an MA model with NAs in your data or sample breaks either.
Note that I say "Most". AR estimation is the notable exception.
Most of the time it doesn't make sense to "treat the filtered sample as a continuous one" since the lagged value of October 2001 is September 2001, whether your sample says so or not.
This is simply because the GARCH estimation process relies heavily on the fact that each observation has adjacent observations available. GARCH is a dynamic process, each observation depends upon the immediate previous observations.
Most estimators in EViews that follow a dynamic process like GARCH require continuous samples. For example, you cannot estimate an MA model with NAs in your data or sample breaks either.
Note that I say "Most". AR estimation is the notable exception.
Most of the time it doesn't make sense to "treat the filtered sample as a continuous one" since the lagged value of October 2001 is September 2001, whether your sample says so or not.
Follow us on Twitter @IHSEViews

 Posts: 2
 Joined: Fri Apr 17, 2009 4:14 pm
Re: NAs in GARCH
Conceptually I understand about the need for a continuous sample.
My question probably reflects my far from expert level of knowledge of EViews at this stage. Since in daily financial data there are going to be quite a number of nontrading days or holidays, is there a more elegant solution for dealing with missing values than cleaning the data by hand and then treating the resulting dataset as an unstructured one in order to run GARCH procedures? Thanks.
My question probably reflects my far from expert level of knowledge of EViews at this stage. Since in daily financial data there are going to be quite a number of nontrading days or holidays, is there a more elegant solution for dealing with missing values than cleaning the data by hand and then treating the resulting dataset as an unstructured one in order to run GARCH procedures? Thanks.

 Fe ddaethom, fe welon, fe amcangyfrifon
 Posts: 11592
 Joined: Tue Sep 16, 2008 5:38 pm
Re: NAs in GARCH
The most elegant solution is to not have the nontrading days in your workfile, by structuring it as dated with date series.
Follow us on Twitter @IHSEViews

 Posts: 2
 Joined: Mon May 04, 2009 11:57 pm
computing continues compound returns
Hi, i am trying to compute continuesl compound returns in eviews but it gives the message of missing values as you klnow stock data does not have regular frequency what should i do.
Re: NAs in GARCH
QMS Gareth wrote:The most elegant solution is to not have the nontrading days in your workfile, by structuring it as dated with date series.
please, how can one structure it as dated with date series? 1OQ.

 Fe ddaethom, fe welon, fe amcangyfrifon
 Posts: 11592
 Joined: Tue Sep 16, 2008 5:38 pm
Re: NAs in GARCH
Open your workfile, click on Proc>Structure/Resize Current Page, and then select Dated with ID Series as the structure type.
Follow us on Twitter @IHSEViews
Re: NAs in GARCH
Hello,
i got the same Problem, i´ve been trying to estimate an GARCH Equation with several Dummies for different Events.
My Problem is, that Eviews only shows "NA" instead of the probability or the standard deviation.
I already tried to resize the page, but i only get the notification "Date Series has no duplicates or gaps. Converting to regular frequency".
Whne i use a single Dummy for all the Events, it works, just the Single Event Dummies won´t.
Do you have any advice?
Thanks in advance.
i got the same Problem, i´ve been trying to estimate an GARCH Equation with several Dummies for different Events.
Code: Select all
'Step2 Estimation USA SingleEvents
equation step2single.arch usa_ind c usa_ind(1) ssr(1) step1_resid tap1 tap2 tap3 tap4 tap5 tap6 tap7 tap8 tap9 tap10 tap11 @ step1_resid(1)^2 tap1 tap2 tap3 tap4 tap5 tap6 tap7 tap8 tap9 tap10 tap11
My Problem is, that Eviews only shows "NA" instead of the probability or the standard deviation.
I already tried to resize the page, but i only get the notification "Date Series has no duplicates or gaps. Converting to regular frequency".
Whne i use a single Dummy for all the Events, it works, just the Single Event Dummies won´t.
Do you have any advice?
Thanks in advance.

 Fe ddaethom, fe welon, fe amcangyfrifon
 Posts: 11592
 Joined: Tue Sep 16, 2008 5:38 pm
Re: NAs in GARCH
I believe you are talking about something different from the original post. The original post is discussing NAs in the data used to estimate GARCH. I think you are asking about NAs in the covariance estimate.
NAs in the covariance estimate are usually due to nonconvergence or singularities. The equation output probably has more detail.
NAs in the covariance estimate are usually due to nonconvergence or singularities. The equation output probably has more detail.
Follow us on Twitter @IHSEViews
Re: NAs in GARCH
Thanks for your fast reply, you´re right i just found this notification:
WARNING: Singular covariance  coefficients are not unique
Presample variance: backcast (parameter = 0.7)
GARCH = C(16) + C(17)*RESID(1)^2 + C(18)*GARCH(1) + C(19)
*STEP1_RESID(1)^2 + C(20)*TAP1 + C(21)*TAP2 + C(22)*TAP3 +
C(23)*TAP4 + C(24)*TAP5 + C(25)*TAP6 + C(26)*TAP7 + C(27)*TAP8 +
C(28)*TAP9 + C(29)*TAP10 + C(30)*TAP11
Do you have any advice how to solve this problem, im relatively new to eviews.
I have already controlled for the dummy variable trap, but this should not be the problem, since i have 11 Dummy Variables, each for a single event, in a total number of 1283 observations.
WARNING: Singular covariance  coefficients are not unique
Presample variance: backcast (parameter = 0.7)
GARCH = C(16) + C(17)*RESID(1)^2 + C(18)*GARCH(1) + C(19)
*STEP1_RESID(1)^2 + C(20)*TAP1 + C(21)*TAP2 + C(22)*TAP3 +
C(23)*TAP4 + C(24)*TAP5 + C(25)*TAP6 + C(26)*TAP7 + C(27)*TAP8 +
C(28)*TAP9 + C(29)*TAP10 + C(30)*TAP11
Do you have any advice how to solve this problem, im relatively new to eviews.
I have already controlled for the dummy variable trap, but this should not be the problem, since i have 11 Dummy Variables, each for a single event, in a total number of 1283 observations.
 Attachments

 test single garch.wf1
 (609.41 KiB) Downloaded 7 times

 test single garch.prg
 (616 Bytes) Downloaded 7 times
Who is online
Users browsing this forum: No registered users and 6 guests