[EMAIL PROTECTED] (Pradyumna S Upadrashta) wrote in message news:<[EMAIL PROTECTED]>...
> >alok wrote:
> >Is there any good test which can figure out if there is an 
> >increasing trend in a set of time ordered data.  Currently 
> >using somthing like a Pairwise Comparision Test (PCT) , but 
> >was wondering if there is anything more robust. PCT =  
> >Summation ( I (Dk > Dk-1) )) / T where T is the total number 
> >of points and I returns 1 if condition is true and 0 if 
> >condition is false.
> 
> Couldn't you simply fit a polynomial and see if you get reasonable fit?
> It's typical in ARIMA modeling to remove trends, seasonality, and so on,
> before fitting an ARMA model.

We at AFS have developed tests for detecting breakpoints in trends as
part of our Intervention Detection proceedures. Please download
http://www.autobox.com/freef.exe and play with our Freeware version
....appropriately named FreeFore .... although some here at
AFS weree considering the name ForePlay which was rejected for
politically correct reasons.

For example consider http://www.autobox.com/trndchange.jpg where a 
statistically significant break point in trend is detected at MAY
2005.
By the way this is monthly prescriptions written in the US for
Diabetes.

Best Regards

Dave Reilly
Automatic Forecasting Systems
http://www.autobox.com
215-675-0652 if you want to chat ....



 




> 
> .
> .
> =================================================================
> Instructions for joining and leaving this list, remarks about the
> problem of INAPPROPRIATE MESSAGES, and archives are available at:
> .                  http://jse.stat.ncsu.edu/                    .
> =================================================================
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to