EViews.com

Hi all,

I’m an EViews7-user.

I have a problem with organizing high-frequency data with a “frenzy” time scale and estimating spline-functions. The data is generated from a mud-core sample taken from the bottom of a lake in Norway. The time-scale is dated by C14-methodology. The problem is that the time frequency is 0.3969 years, i.e. the time difference between two observations is 0.3969 years (or 4.76 months!!). As you see – every 2.5 observation represents a year. Due to odd frequency it is not possible to organize the data so it could be possible to measure the seasonality-effects as we can do by using quarterly or monthly data. The scientific objective in this case is to measure the long-run properties (trend, cycles, structural changes/shifts etc.) in addition to estimating the statistical model.

I have the following question: Is it possible to reorganize the original data – and generate a new data-file, by (a) selecting, for example, every 10th observation (if we select every 25 observation then the time interval (frequency) is 10 years – alternatively (b) calculate the average of, say, every ten or 25 observation, and use these data as the new input file. (c) Let’s say that we want to estimate a spline-function but we do not know where the knots are located. The problem is: How do we estimate spline functions with unknown knots? How do we formulate the problem in EViews?

I appreciate a lot if you have some good ideas how to solve these problems. If you need to look at the original data-file, please let me know.

Regards,
Torbjørn

Well, I'm pretty sure we've never had a request for a .3969 year frequency before!

As far the seasonal adjustment issue goes, the first thing I'd point out is that your problem isn't entirely different from the problem of seasonally adjustment of weekly data, since there are 52.1785 weeks in a year which is also not an integer multiple. You may like to look for discussions of seasonal adjustment of weekly data for ideas on how to handle your problem.

One idea that shows up in this sort of area is the use of trigonmetric cycles instead of period dummies to represent the seasonality. There is nothing stopping you from using trigonometric cycles with a period that is not an integer multiple of the observed frequency of your data, so in theory you could set up trigonometric cycles at annual frequencies even though your data is observed every 4.76 months.

EViews doesn't have much in the way of built in procedures for doing this, but there's nothing really stopping you from trying some of this yourself. As a starting point for ideas, you may like to look at the discussion of seasonality in Chapter 2 of the book "Forecasting, structural time series model and the Kalman filter" by Andrew C. Harvey.

As far as reorganizing the data into ten year intervals, you can certainly do this in EViews. Again, it's not exactly something that is built in for your frequency, but one approach would be to use the @mod() function applied to a trend in the original workfile to make a new identifier series that grouped together observations. You could then use various aspects of EViews match merging of pages to make a workfile with observations at the new frequency.

Estimating splines with unknown knot positions is not an area that I know much about I'm afraid (and not an area where EViews has any built in support). It's not something we've had requests for, although there's almost surely a literature out there on the subject.

I'm afraid you're a long way outside of typical uses for EViews. There's still a fair few general purpose tools in EViews that may be useful to you, but we aren't particularly strong on built in procedures in the areas in which you're working.

Thank a lot for your answer, Chris. Frustrating problem. The paleo-people sent me the data, and I can tell you they operate occasionally with odd measurments - in this case odd frequencies. As you say - maybe a trigonometric approach will do. I have Harvey's book so I will brush the dust of chapter 2 and do my best to understand it. I will also see what I can do by using the the @mod() function.

Merry Christmas to you and your colleaques at EViews.

Torbjørn.