1
$\begingroup$

I am regressing monthly data that I know has significant seasonality. I have about 50 monthly observations. I was thinking of using 12 variables and for each row of my data turn on one variable depending on the month. I was then going to see what monthly variables were significant and then rerun for the these possibly grouping when I see months that appear very similar.

I know there are more comprehensive ways of approaching this but I am only working with Excel.

Improve this question
$\endgroup$
1
  • $\begingroup$ That will work, but you should use only m-1 variables (11 in this case). Variables 1-11 will be all zeroes for December (which is how you will know it is December). Variable 1 will be 1 in January, otherwise 0. Variable 2 will be 1 in February, else 0. Excel isn't ideal, but will work. $\endgroup$ – zbicyclist Jun 8 '13 at 4:16
2
$\begingroup$

I'd recommend moving to R, you'll avoid errors that way. There are many ways you can handle seasonality, and it'll largely be impacted by the trend in the stochastic variable. You can use a binary variable for each month, you can also use a binary for each season (i.e., Spring, Summer, Fall, Winter), or you can use Harmonic Regression (which is much more interesting).

They'll all give you different results and they're all essentially arguments about what the trend is in the data, but I think it makes sense to move towards Harmonic Regression. Harmonic Regression is essentially using sinusoidal waves to decompose the underlying seasonal variation.

http://www.amstat.org/meetings/jsm/2010/onlineprogram/AbstractDetails.cfm?abstractid=307445

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.