Page 2 of 4

Re: Frequency Conversion

Posted: Wed Aug 10, 2011 8:55 am
by EViews Gareth
Could you provide more details?

Re: Frequency Conversion

Posted: Wed Oct 05, 2011 2:09 pm
by jason_ll
Hi there,
I have a few variables in an Excel sheet. They are annual and my workfile is quarterly. Every time I import them, they get converted using the constant match (average) method. I keep trying to change the method (I want it to be linear), but no matter which frequency conversion option I choose, it always gets conveted according to the constant match method.

I even tried changing the default Eviews conversion, but it still doesn't work. It gives me data that looks like this:

2000Q1 30.29750
2000Q2 30.29750
2000Q3 30.29750
2000Q4 30.29750
2001Q1 25.92250
2001Q2 25.92250
2001Q3 25.92250
2001Q4 25.92250

I was only able to fix this problem in two steps. First copying the data into Eviews as it appears above, and then create a series link that converts it using the linear converstion. That did work, but it takes up too much time and I don't want to work with a "series link".

Thank you!

PS: I'm importing the data from Excel using the File -->import --> import from file method.
In step 3 out of 3, you get the option of choosing the conversion method. Btw, conversion has always worked flawlessly from high to low frequency, but now it doesn't seem to work the other way around.

Re: Frequency Conversion

Posted: Wed Oct 05, 2011 2:17 pm
by EViews Gareth
Importing and performing frequency conversion only supports a couple of conversion methods from low to high. The dialog shows the full list of options available, but only constant match actually work :(

Re: Frequency Conversion

Posted: Thu Oct 06, 2011 6:03 am
by jason_ll
EViews Gareth wrote:Importing and performing frequency conversion only supports a couple of conversion methods from low to high. The dialog shows the full list of options available, but only constant match actually work :(

That's strange. Thanks for the confirmation.

Re: Frequency Conversion --> Cross-section frequency convers

Posted: Fri Dec 02, 2011 6:43 am
by ewijks
Hi Gareth,

I am working with a large panel dataset, in which I need to perform a high to low frequency conversion, not in the time dimension, but in the cross-section.

Thus, one workfile page has the following dimensions: 2001 2010 x 111,000 or around 800,000 observations
The cross-section identifier here is off, which is a unique number for specific offices. The page also contains a series called inst, which matches the offices with the institutions to which they belong.

The other workfile page has the following dimensions: 1992 2010 x 16,000 or around 190,000 observations.
The cross-section identfier here is inst ,and corresponds to inst in the first workfile page.

I need to paste series from the first page in the second page.
Date should just be matched (both workfile pages are annual, the only difference is that the date series of first workfile page contains June dates, e.g. 2003-06-30, and the date series of the second workfile page contains year-end dates, e.g. 2003-12-31), but there should be a conversion of the cross-section from high (off) to low (inst) frequency.
Key to matching the data correctly is inst, which is contained in both workfile pages.

Is this possible?
And if yes, how do I do it?

I am looking forward to your response, thank you in advance.

Best regards,
Saskia

Re: Frequency Conversion

Posted: Tue May 29, 2012 3:18 am
by atoch
Bonjour,

Thank you for this post on frequency conversion.
Setting them up seems very easy either with the link or copy commands.
I'm looking for an instruction to extract the conversion method used from objects that already exist in a workfile.
Ie: in an annual page, GDP is a link to a quarterly page. What is the command to retrieve if annual GDP was build as a sum or an average ?

Thanks.

Re: Frequency Conversion

Posted: Tue May 29, 2012 7:42 am
by EViews Gareth
If it is a link, you can open up that link, hit the "Properties" button, and then look at the "Link Spec" tab to see which conversion method is being used.

Re: Frequency Conversion

Posted: Wed Jun 27, 2012 2:29 am
by Fenix
Hey!

Is there a way to have linear interpolation without using the last ovservation, just use average values?

I have annual data. I convert it to quarterly. If I choose constant-match average and paste it as a link I get just the annual values divided by four in the spreadsheet BUT when I graph it, I get a linear interpolation in the graph. I want exactly the linear interpolation. If I unlink the series, the graph changes to the stepwise one. I.e. is there a way to export the data from the graph (even thought it says it uses the Raw Data, it does not).

I tried the 'linear - match last' but it doesn't work for me.

More details: I have annual series for total population from 1999 to 2011. Year 1990 is 470'388'255. I want a simple linear interpolation, where the average of the values 1990Q1 to 1990Q4 is 470'388'255.
If I use " constant-match average " I get
1990Q1 = 470'388'255
1990Q2 = 470'388'255
1990Q3 = 470'388'255
1990Q4 = 470'388'255
BUT the graph does not look stepwise as long as I have the series as a link. So if I have a link, the graph looks like that what I want in the end but do not know how to get the data.

If I use 'linear - match last', then this is true for the average of 1990Q4 - 1991Q3 and in the end I am missing 3 observations, since the value for 2012Q1 is the value for the whole 2011.

Re: Frequency Conversion

Posted: Wed Jun 27, 2012 7:52 am
by EViews Gareth
No, you cannot do that.

Re: Frequency Conversion

Posted: Sat Jul 21, 2012 1:15 pm
by iftikharferoze
I tried to do data converting from quarterly to monthly, but in the first two row of my data set was vanish. I just could see NA expression in that rows. what I need to do to get rid of this. I use version 5

Re: Frequency Conversion

Posted: Sat Jul 21, 2012 1:24 pm
by EViews Gareth
Hard to say without seeing the data.

Re: Frequency Conversion

Posted: Wed Aug 08, 2012 1:07 pm
by mylo47
Hi
I noticed the following discrepancy in the conventions for frequency conversion. When I convert data from daily to monthly using, say, option c=l, EViews chooses the value from the last day of the given month. This is logical. However, when I convert data from daily to weekly (again, with option c=l), then EViews chooses the value from the last day of the following week, not of the given week. This is inconsistent with the monthly convention and is fraught with the danger of look-ahead bias in forecasting work. Is there a way to select the value from the given week in a systematic way? Also, any ideas what was the thinking behind this design?
Thank you

Re: Frequency Conversion

Posted: Wed Aug 08, 2012 1:50 pm
by EViews Chris
In all cases, 'last' should return the last high frequency observation that falls within the lower frequency observation.

My guess is that what is confusing you is the definition of the week. The week begins on whatever day you use as a start date and ends on the seventh day counting from that. EViews does not force weeks to start on any particular weekday.

You can use

show @strdate("weekday")

to see what day of the week each observation of your weekly workfile begins on.

And you can use

show @date @enddate

to see the range of dates within any observation.

Does that explain things?

Re: Frequency Conversion

Posted: Thu Aug 09, 2012 6:08 am
by mylo47
Chris
Thanks a lot for the explanation - this clears up the concept. Is there any way to redefine the week as the 7 days before a given day?
Cheers
George

Re: Frequency Conversion

Posted: Thu Aug 09, 2012 9:21 am
by EViews Chris
You'd have to shift the start date of the weekly workfile back by six days.

In a program you could do something like this:

series dateshift = @dateadd(@date, -6, "day")
pagestruct @date(dateshift)
delete dateshift

EViews always labels weekly observations with the beginning-of-period date in the spreadsheet view (the same value as the implicit series @date). We could provide an option for displaying end-of-period date in this view as well (the implicit series @enddate) - although I suppose that might confuse people even more if they weren't sure which one they were looking at. The unambiguous thing would be to display the full range from start date to end date all the time but that would take up a lot of screen real estate. Providing @enddate as an implicit series is a sort of compromise - it's there if you want to see it.