Calculates future values based on curve-fitting to historical values. The @TREND procedure considers a number of observations; constructs a mathematical model of the process based on these observations (that is, fits a curve); and predicts values for a future observation. You can use weights to assign credibility coefficients to particular observations, report errors of the curve fitting, choose the forecasting method to be used (for example, linear regression), and specify certain data filters.
@TREND (Ylist, [Xlist], [weightList], [errorList],
[XforecastList],
YforecastList, method[, method parameters] [, Xfilter1
[, parameters]] [, XfilterN [, parameters]]
[, Yfilter1 [, parameters]] [, YfilterN [, parameters]])
Ylist | An expression list that contains known observations; for example, sales figures over a period of time. |
Xlist | Optional. An expression list that contains underlying variable values. For example, for each sales figure in Ylist, Xlist may contain a value for associated time periods. If you do not specify Xlist, the default variable values are 1,2,3, and so on, up to the number of values in Ylist. |
weightList | Optional. An expression list that contains weights for the data points
in Ylist, for the linear regression method only. If values in
weightList are #MISSING , the default is 1. Weights
for methods other than linear regression are ignored. Negative weights
are replaced with their absolute values. |
errorList | Optional. Member list that represents the differences between the data points in Ylist and the data points on the line or curve (as specified for method). |
XforecastList | Optional. Expression list that contains the underlying variable values
for which the forecasting is sought. If you do not specify XforecastList,
the values are assumed to be as follows: {(last value in Xlist + 1), (last value in Xlist + 2), ...} up to (last value in Xlist + the number of values in YforecastList) If you forecast consecutively from where Ylist stops, you do not need to specify XforecastList. If you want to move the forecasting period forward, specify the new period with XforecastList. |
YforecastList | A member list into which the forecast values are placed. |
method | A choice among LR (linear regression), SES (single exponential smoothing),
DES (double exponential smoothing), and TES (triple exponential smoothing).
Method parameters must be numeric values, not member names. Method parameters
may be any of the following:
|
Xfilter1 ... XfilterN | Optional. Use one or more of the following filter methods to scale
Xlist:
|
Yfilter1 ... YfilterN | Optional. Use one or more of the following filter methods to scale
Ylist:
|
#MISSING
values during calculation
of the trend.#MISSING
values.
Also, if you specify Xlist, the data must be equidistant, with
the interval (step) being a whole fraction of the period, T (for example,
T/5, T/2). The XforecastList parameters should also contain multiples
of the interval.The following example is based on the Sample Basic database. It forecasts sales data for May through December, based on the trend of the same sales data from January through April. The method used is linear regression with no seasonal adjustment.
Sales(@TREND(Jan:Apr,,,,,May:Dec,LR););
This example produces the following report:
Actual Sales West Colas ===== Jan 2339 Feb 2298 Mar 2313 Apr 2332 May 2319 Jun 2318.4 Jul 2317.8 Aug 2317.2 Sep 2316.6 Oct 2316 Nov 2315.4 Dec 2314.8 Year 27817.2
Click the following links to see the algorithms used for each method:
©2004 Hyperion Solutions Corporation. All Rights Reserved. http://www.hyperion.com |