On Fri, 10 Sep 2010, Anton Bletsov wrote: > I am relatively inexperienced with Gretl, and I had a problem while > attempting to develop a time series model with ARIMA. In particular, once I > gave the parameters and chose the variables for my ARIMA, (dependent + 9 > independents, 15 observations; AR order = 1, Difference=1; MA order =1), I > got the following error message: "BFGS: initial value of objective function > is not finite". > > Is there something I can do to resolve this problem?
Yes, add another hundred or so observations! Seriously, Maximum Likelihood ARIMA estimation is based on asymptotic considerations, and it is hardly suprising that you ran into numerical problems with a sample size of 15, which would generally be considered too small for any sort of statistical analysis, let alone ARIMA. Allin Cottrell
