Aggregate forecasts of shorter intervals for longer intervals

[Chichi Shu]

Hi, Listers,

 

I have a series of weekly data but I want to get the monthly forecasts.

 

So is it valid if I forecast on a weekly basis and aggregate the weekly forecasts for the future months’ forecasts?

 

Or generally speaking, is result of aggregating forecasts on a shorter interval for getting forecasts on a longer interval valid comparing to aggregating the original data of shorter intervals first and then forecasting on that aggregated longer interval?

 

Is there any literature that I can refer to about this problem?

 

Thanks!

 

 

[Jon K Peck]

Post aggregation is perfectly okay, but either method will do.  You have to decide on the appropriate aggregation method - usually sum or average depending on the nature of the data.  You also have to decide on how to aggregate weeks into months, since they do not fit exactly in either case.

The reverse problem, disaggregating monthly forecasts into weekly forecasts, is harder, but there are methods for that as well.


Jon Peck (no "h") aka Kim
Senior Software Engineer, IBM
[hidden email]
phone: [phone]




From:        Chichi Shu <[hidden email]>
To:        [hidden email],
Date:        06/20/2014 11:29 AM
Subject:        [SPSSX-L] Aggregate forecasts of shorter intervals for longer              intervals
Sent by:        "SPSSX(r) Discussion" <[hidden email]>

---

Hi, Listers,
 
I have a series of weekly data but I want to get the monthly forecasts.
 
So is it valid if I forecast on a weekly basis and aggregate the weekly forecasts for the future months’ forecasts?
 
Or generally speaking, is result of aggregating forecasts on a shorter interval for getting forecasts on a longer interval valid comparing to aggregating the original data of shorter intervals first and then forecasting on that aggregated longer interval?
 
Is there any literature that I can refer to about this problem?
 
Thanks!
 
 

[Chichi Shu]

Thanks, Jon.

 

I don’t think there is a problem with this approach in terms of the point estimator of the weekly forecasts either. But if we look at the upper bound and lower bound of 95% confidence of a fiscal month (aggregated weekly) forecast, doesn’t this kind of adding up significantly increase the variance? i. e. Assuming the forecasting model successfully produces homoscedastic weekly forecasts, the four week aggregation gives us an 16 times as big as the individual week’s variance?

 

From: [hidden email]

Sent: Friday, June 20, 2014 4:29 PM

To: [hidden email]

Cc: [hidden email]

Subject: Re: [SPSSX-L] Aggregate forecasts of shorter intervals for longer intervals

 

Post aggregation is perfectly okay, but either method will do.  You have to decide on the appropriate aggregation method - usually sum or average depending on the nature of the data.  You also have to decide on how to aggregate weeks into months, since they do not fit exactly in either case.

The reverse problem, disaggregating monthly forecasts into weekly forecasts, is harder, but there are methods for that as well.


Jon Peck (no "h") aka Kim
Senior Software Engineer, IBM
[hidden email]
phone: [phone]




From:        Chichi Shu <[hidden email]>
To:        [hidden email],
Date:        06/20/2014 11:29 AM
Subject:        [SPSSX-L] Aggregate forecasts of shorter intervals for longer              intervals
Sent by:        "SPSSX(r) Discussion" <[hidden email]>

---

Hi, Listers,
 
I have a series of weekly data but I want to get the monthly forecasts.
 
So is it valid if I forecast on a weekly basis and aggregate the weekly forecasts for the future months’ forecasts?
 
Or generally speaking, is result of aggregating forecasts on a shorter interval for getting forecasts on a longer interval valid comparing to aggregating the original data of shorter intervals first and then forecasting on that aggregated longer interval?
 
Is there any literature that I can refer to about this problem?
 
Thanks!
 
 

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD