Cosine with drift in time series.

Hi!

I am currently working with some electricity system price data.

The data exhibits quite a bit of seasonality and drift. I need to capture this to build my model. I want to use a cosine for the seasonality and a linear function to capture the drift.

Does anybody have any tips for how i can implement this? For most I need some articles or books that explain how to implement this. But if anybody has some tips for code, this would also be very useful.

I have seen that the business cycle literature speaks of the decomposition of trend and cycle, so if anybody has some tips that include liner drift and a cosine, that would be great. Tried the HP filter, but its too rough.

Hope you got any tips!

Best Martin

Answer 1

[the cyclist]

I suggest you peruse this site:

http://www.spatial-econometrics.com/

[Martin Gullaksen]

I have worked a bit with LeSages stuff before, but I am a bit unsure where too look. Can you please be a bit more specific?

[the cyclist]

Well, not really. I used that toolbox years ago, when I was working with some economists, and I recall it handled issues you are talking about. However, I don't remember the specifics. I thought maybe you could poke around the site and/or do some keyword searching there.

Sorry I can't be more helpful than that.

[Martin Gullaksen]

Ok, I have started reading LeSage's book. Hope I find something.

Thanks for the tip anyway. Its a great toolbox.

BTW: what happens when you accept an answer in this forum?

[Walter Roberson]

When you accept an answer, it gets visually marked as accepted, and the person who posted the answer is given 4 reputation points and the number of "accepted" answers is increased by 1 for them.

Answer 2

[hans]

'fminsearch' might be a handy tool.

With it You can choose a type of function that You would like extract and calculate the coefficients.

function
    ...
    AStart     = [1 2]
    ABest      = fminsearch(@(AStart) QualFun(AStart,x,P),AStart);
    ...
end

function [LSqr]=QualFun(A,x,P)
    fit   = A(1)*x+A(2);
    Diff  = fit-P;
    LSqr  = sum(Diff.^2);   
end