Re: [Gretl-devel] constant mis-specification and arima convergence problem

I mean, there are 2 frequent reasons for current arima non-convergence: 1) redundant constant indicators: slowly growing lnl and small gradients large scaling factor (as is seen using --verbose p-value for constant is close to 1 when convergence is achieved by some tricks or by --x-12-arima (some tricks are in the script) The problem is aggravated with seasonal over-differencing (e.g. D=1 instead of D=0) 2) missing constant indicators: some indicators: large norm of gradient as a direct cause of non-convergence nearly unit circle pol. roots when convergence is achieved by some tricks or by --x-12-arima (some tricks are in the script) Very low p-value for const in model with the same p d q; P D Q Hopefully, this info (along with the tricks in the script) may help to make some modifications for 197 Oleh 22 жовтня 2018, 08:59:50, від "Riccardo (Jack) Lucchetti" <r.lucchetti(a)univpm.it&gt;: