Thanks for this, extremely helpful. I am getting this to work with lsq first and then I'll tackle arma. I am approaching the library in this way as I need to run estimation on a window of data, and then increment forward throughout a time series period, with a moving window. So, for each period t, I need to estmate the residuals from the previous X days, based on a regression from the same previous x days. On day t+1 is another regression, etc. If you have a preferred way for me to do this, certainly let me know. Perhaps I should look into writing scripts and using issuing gretlcli commands, as it's a batch-like process. I have it all working with OLS at the moment, I can get residuals and it matches gretl. The only thing missing is the coefficients of the constant and independent variable. Having a working OLS model, these are the additional pieces I added (same result using matrix): gretl_vector *coeff=NULL; // same results using gretl_matrix / * model runs */ coeff = gretl_model_get_matrix(model, M_COEFF, errcode) The function returns NULL, unfortunately, with a very large error code. Is a specific option required to get this matrix return? The scalar M_NCOEFF returns 2 as expected. This is hopefully the last thing I need. Thanks! Joel --Forwarded Message Attachment-- From: cottrell(a)wfu.edu To: gretl-users(a)lists.wfu.edu Date: Sat, 3 Jul 2010 13:48:49 -0400 Subject: Re: [Gretl-users] Using Gretl API for C++ Program On Wed, 30 Jun 2010, Joel Bycraft wrote: Sorry this has taken a while. For general information on how gretl "lists" work see the "Lists" sub-section of the libgretl API docs (under the "Utilities" heading). Note that I've done quite a bit of work on these docs lately, though they're still very much incomplete and are unlikely to be completed any time soon. http://gretl.sourceforge.net/API/ You should refer to the help entry for the "arima" command in gretl to understand this. The specification of the arma list is given in detail there. (In general, where the API docs are obscure on a given function, and that function directly implements a gretl command, that's what you'll have to do.) Although there are other ways of doing this, for the present you're probably best constructing the list argument for arma by composing a string as per the help on arima, then calling gretl_list_from_string to convert it to an actual "list". For example: <C-code> const char *arma_spec = "1 2 ; 5"; int err = 0; int *list = gretl_list_from_string(arma_spec, &err); /* if err == 0, use the list ... */ free(list); /* once you're finished with it */ </C-code> In this example the AR order is 1, the MA order is 2, and the dependent variable is at position 5 in the "Z" data array. To add seasonal terms, again look at the "arima" help. The assumption in libgretl is that a two-dimensional data array Z (of type double **) lives somewhere in the calling program. With some libgretl functions it is sufficient to pass Z as an argument result = some_gretl_func(..., Z, ...); but with others you must pass its address: result = some_other_func(..., &Z, ...); The functions that require the address of Z are ones that may resize the Z array under some conditions: lsq is one such (it may add auxiliary series to the data array temporarily). The arma function, on the other hand, never resizes Z so it doesn't require the address. Note that whenever you see "***pZ" as a libgretl function parameter, the corresponding argument should be "&Z". (Well,of course you can call "Z" anything you like in your own program.) Just out of interest, do you have a specific reason to do ARMA modeling "the hard way", using libgretl yourself, rather than using gretl? Allin Cottrell http://www.econ.univpm.it/lucchetti _________________________________________________________________ Game on: Challenge friends to great games on Messenger http://go.microsoft.com/?linkid=9734387