GROWTH too aggressive, TREND too conservative

GROWTH too aggressive, TREND too conservative - what to do?

Thread view:

Enable EMail Alerts

Start New Thread

Thread rating:

Brett L. - 15 Mar 2007 02:12 GMT

I am working on projecting some data that is non-linear. the TREND function
gives me a line that is too conservative in its estimates, but GROWTH gives
me an exponential line that is way too aggressive.

Is there another function I can use?
Can I combine these functions meaningfully?
Is there a way to tweak GROWTH to be less aggressive?
Any other ideas?

Reply to this Message

Harlan Grove - 15 Mar 2007 04:42 GMT

Brett L. <Brett L...@discussions.microsoft.com> wrote...

>I am working on projecting some data that is non-linear. the TREND
>function gives me a line that is too conservative in its estimates,
>but GROWTH gives me an exponential line that is way too aggressive.

Idle curiosity: which gives the higher R-squared value?

>Is there another function I can use?

Assuming you have one y series and one x series, there are other
functional forms you could try. General ones,

Polynomials: y = a + b x + c x^2 + d x^3 + . . .

=LINEST(y,x^{1,2,3,...})

Hoerl curves: y = a x^b exp(c x)

=LINEST(LN(y),LN(x)*{1,0}+x*{0,1})

There's also logistic regression.

=LINEST(LN(y/(1-y)),x)

>Can I combine these functions meaningfully?

You could always average the two, but that'd be pure ad hoc without a
shred of theoretical support. Even so, it may serve your needs
adequately.

>Is there a way to tweak GROWTH to be less aggressive?

Not without transformingadjusting either x or y series.

Reply to this Message

Jerry W. Lewis - 15 Mar 2007 12:30 GMT

...

> Assuming you have one y series and one x series, there are other
> functional forms you could try. General ones,
>
> Polynomials: y = a + b x + c x^2 + d x^3 + . . .
>
> =LINEST(y,x^{1,2,3,...})

...
which can be projected by TREND as

=TREND(y,known_x^{1,2,3,...},new_x^{1,2,3,...})

Jerry

Reply to this Message