Why is the "included observations" reduced

[JimS]

I am transferring from econometrics coding in TSP and RATS to Eviews and to help me learn the differences I am trying to replicate the output from the in-built ARCH routines using the ml command on a logl written in a .prg. I have used the posted trivariate GARCH in mean to run a univariate GARCH(1,1) in mean (plus just a constant in the mean equation). I use the in-built ARCH routine to set starting values and get a close but not exact replication using my ml logl. The reason seems to be that the ml logl is dropping the first year (252 observations) of my dataset (2362 obs), while the inbuilt routine drops only the first observation. My sample statements request the full sample less the first couple of observations as in the posted example program. When I extend the model by adding an AR(1) and exogenous variables in the mean, the same thing happens. But, when I also add exogenous variables to the variance equation, the ml logl drops a further year of observations (504 observations in total). So now the results from the inbuilt routine and my ml logl are substantially different. So, my question, can I stop the ml logl from dropping observations and, if so, how? Thank you.

[EViews Gareth]

The built in routines should drop the same number of observations as your LogL object, assuming you did everything correctly.

[JimS]

Thanks for the prompt reply, Gareth, but I'm none the wiser. I'm aware from other programming environments that these problems are usually related to how one sets up the sample period, but I think I've followed the examples properly. Anyway, I've attached the program file and the two estimation outputs (that differ by one year of included observations), in the hope that you can spot what I'm not doing correctly. I'm using Eviews 8. Thank you.