Re: [Gretl-devel] Bug in NLS?
On Wed, 17 Aug 2011, Marcin Błażejowski wrote:
simple script:
<hansl>
open greene11_3.gdt
# OLS block
ols C 0 Y
fcast ols1
fcast ols2
fcast ols3
# NLS block
genr a = 1.0
genr b = 1.0
nls C = a + b * Y
params a b
end nls
fcast nls1
fcast nls2
fcast nls3
</hansl>
All 3 forecast generated by OLS (olsX) are the same while forecast generated
by NLS (nlsX) are different. Moreover, olsX = nls1 (always, no matter the
number of forecast is).
Yes, the nls forecast was wandering on repetition of the
fcast command; thanks for the report.
This was due to "experimental" code which added a bootstrapped
forecast error: the trouble was that the original parameter
values were not being restored after the bootstrap. That's now
fixed in CVS and snapshots.
Allin