Seasonal Forecasting Trend

I'm trying to figure out how to do seasonal forecasting based on volumes. I tried to use FORECAST but the numbers are not accurate.

I have actual data to work with to come up with my forecasting on a monthly basis. Our yearly calender is from Nov 08 to Oct 09.

Here's a sample of the data.

Nov-08 Dec-08 Jan-09 Feb-09
116,529 116,579 115,990 116,478
1,197 1,212 1,261 1,255
1,021 713 985 1,268

How do I do my forecasting for Mar-09? It has to be seasonal differences.

Any help is appreciated. Thanks.

You only have four months of data -- how could you extract seasonal variation from that? Five or ten years worth would be a good start.

And, without providing the data, you'd like a seasonal estimate?

Hit Reply, push the button with the paper clip (or scroll down to the Manage Attachments button), browse to the file, press Upload.

This may be correct, but I couldn't swear to it. I see very little seasonal variation in the data.

The line below the data takes the ratio of the month to the yearly average excluding the month. (Literature says to compute two averages, and then average them: six months before plus five months after, and five months before plus six months after. The filter I use does that in simpler fashion.)

The line below that normalizes the average about 1, i.e., the average of the second row is exactly one. (It would be better if there were more than one whole year of data).

The forecast uses an underlying exponential fit (the GROWTH function) to forecast the growth, corrected by the seasonal variation computed above.

I think the whole thing would be accurate if you had an integer number of years (several, four or five) over which to compute the seasonal variation. As it is, the seasonal variation itself affects the exponential fit -- but again, there ain't much seasonal variation.

If you work for the Fed, please consult an expert

You’re welcome. Would you please mark the thread as Solved?

Click the Edit button on your first post in the thread

Click Go Advanced, select [SOLVED] from the Prefix dropdown, then click Save Changes

For deriving a seasonal (monthly) factor, you want a centered average. That would be easy if a year had 11 months, an odd number; you would average the values from five previous through five following.

Inconveniently, a year has 12 months, an even number. So instead, you average 13 months, with half the weight applied to the months at either end.