Re: [Gretl-users] Regression with ex-ante boundaries imposed on forecasted values
Am 24.02.2015 um 13:53 schrieb Daniel Bencik:
Dear forum,
this is more of a question related to econometrics. When, for
example, your goal is to model/forecast weekly highs/weekly lows,
when you run your regression on Tuesdays, you already know that the
model should not predict a weekly high below the Monday's high. The
Wednesday's prediction of the whole week's high should not be below
max(mondayHigh, tuesdayHigh). My questions is whether there is an
econometric tool/approach that is capable of estimating a model
bearing this in mind. That is, I want the estimated coefficients to
take into account, that the forecasts should not be below/above some
value which changes over time (i.e. it is not a constant like e.g.
zero or something).
If I understand your question correctly, there is a trivial solution,
although you may not like it: Produce forecasts with standard tools, and
then apply your time-varying max() operator.
Or you could specify your model in terms of squared deviations (or the
negative of that) centered on your previous high, and your restriction
would hold. I'm not sure that makes much sense, but econometrically it's
not a big deal unless you also want to have some other optimality
properties.
cheers,
sven