LINEST also returns #REF! when you exceed its hardcoded limit on
predictors. Empirically, that limit appears to be 17 predictors if no
constant is fitted, or 16 predictors and a fitted constant.
As a practical matter I would be very suspicious of the numerical
properties of LINEST solutions long before hitting that hard coded
limit. Perhaps that is why it is there, even though MINVERSE in theory
can invert a 52x52 X'X matrix.
Jerry
Harlan Grove wrote:
Peter N wrote...
I am using Excel 2002. It seems to be unable to handle any more than
16
predictor variables. All it returns is a #REF error. The help does
not
document any limit. Is this an undocumented limitation? Is Excel
2003
LINEST capable of handling more that 16 variables? How many?
What's your *EXACT* formula? LINEST should return #VALUE! and #NUM!
errors when it can't invert the bilinear form of the independent
variables, but it only returns #REF! when there's a true range
reference error.
|