can anyone help? I have a small group project looking to do seasonality forecast. The problem provides sales figures for 3 years in monthly increments (Jan.-Dec. year 1, year 2, year 3), and we're asked to do the seasonal indexes for each month and show adjustments. I tried doing it with 4 seasons but couldn't get the periods to setup right with the 4 period moving averages. And without the correct setup, I can't get the the index numbers and the deseaonalized figures. I also tried 3 seasons (figuring 3 months per season?) and the numbers were all messed up.

any suggestions on how to setup the problem? Any help would be greatly appreciated.

 

you are not using enough widgets and thingamadooies. by adding more widgets and thingamadooies (and a thingamajig and doohickey) the equation will be properly set up. hope this helps............

 

[QUOTE]Originally posted by spike
[B]you are not using enough widgets and thingamadooies.

 

Hi Mingster,

How about taking three month sales averages for the 4 seasons of each year and then do this for a total of 12 seasons (3 years).

Now look at the % change from season to season during each single year.

I would expect you will see a consistant trend (ex 15% decrease in sales for winter).

The problem then may ask you to predict sales for a particular season in year 4 given some growth factor.

Have fun!

 

Thanks!

But the question asks for the seasonal indexes for each month - would averaging 3 months of sales neglect to answer that? the graph we charted out with the data shows a consistent seasonal trend as you've stated.

 

My recollection is that it goes something like this:

Compute the sales average for January (the average of the sales for Jan 0001, Jan 0002, and Jan 0003). Do the same for February, March, . . ., December. Call these monthly averages m1 (Jan), m2 (Feb), . . ., m12 (Dec).

Compute the total sales average for all 36 months. Call this overall average a.

The January index will be i1 = m1/a, the February index will be i2 = m2/a, . . ., the December index will be i12 = m12/a.

Divide all of the January sales figures by i1 to remove the seasonality. Do the same for February (divide by i2), March (i3), . . ., December (i12).

Use these adjusted sales numbers for your forcasting; e.g., you might compute a regression line and use that line for future predictions.

Finally, adjust the future predictions by the appropriate monthly index to get the seasonally-adjusted prediction. For example, use the regression line to get a (seasonally unadjusted) prediction for January 0004, then multiply this number by i1 to get the seasonally adjusted prediction for January 0004. This last number is the one you would report.

At least, that's how I remember it.

Hope this helps.



PS If you need it by "season" instead of month, you use the same scheme, just lump the months together into seasons. For example, you might call December, January, & February "Winter", March, April & May "Spring", June, July & August "Summer", and September, October & November "Autumn". (It might be easier to go JFM, AMJ, JAS, OND given the data you have.) Then you'll have quarterly averages, an overall average, and quarterly indices. You'll still base your (unadjusted) prediction on the deseasonalized (divided) quarterly numbers, then adjust them with the quarterly indices.

If you like, I can e-mail you a spreadsheet with an example.

 

It sounds like they're using "season" to mean "month". If so, do what I suggested before the PS.