Forecasting based on historical data & YTD actuals

[solo9999]

Hi all,

Kind of in a pickle here! I have the following scenario:

Jan-Dec 2010 data by the month (in let's say USD per month)
123,110,139,166,178,138,137,178,172,151,110,115

Jan-Aug 2011 data for the same
165,145,159,144,103,125,103,138

Can I use the above to forecast Sep-Dec this year? There is seasonality invloved which is reflected in the 2010 numbers. The issue is that for example Jan-Mar 2010 is lower than Jan-Mar 2011 and then it switches around. My thoughts on this is that it must be a formula that will calculate the difference between 2010 & 2011 month-to-month Jan through Aug but it should also take into consideration the 2011 trend. Is that even possible?

!!!!Thanks!!!!

 

[doubledaffy]

The month (seasonality) of the data doesn't seem to be a very good predictor of data you want. You could come up with a regression, but it wouldn't be very accurate. It's R-squared value would only be around 0.3

 

[solo9999]

Thanks. I think I will try to get more data going back to 2008 and base it on that. Oh and I have to learn more about forecasting

 

[doubledaffy]

More data might help, but the low R-squared value indicates that the values depend not just on the month, but something else. So, without finding out what that (or those) other variables are, you can't make a good prediction for future values.

That said, you can still do a forecast with the low R-squared fit I mentioned before, and use the values. Since we know that it isn't a good fit, you can use the values as a best guess with the understanding that the real value could be quite different. Not ideal, but it's better than nothing, and isn't fallacious.

 

[solo9999]

More data might help, but the low R-squared value indicates that the values depend not just on the month, but something else. So, without finding out what that (or those) other variables are, you can't make a good prediction for future values.

That said, you can still do a forecast with the low R-squared fit I mentioned before, and use the values. Since we know that it isn't a good fit, you can use the values as a best guess with the understanding that the real value could be quite different. Not ideal, but it's better than nothing, and isn't fallacious.

Yes, the data is cost figures in a volatile industry There are several one-time effects in there that also influence the figures. I think I will try to obtain more data and play with a regression to see how close the forecast will be to reality. Even if it gives me a basis that I can manually adjust as time goes it will still be good

Thanks!