On Tue, 12 Jun 2007, Chris wrote: The arch command is a bit unusual (this should be probably be changed): it's a test, plus ARCH estimation _if_ the test if significant. I suspect that in the case you're talking about the ARCH effect is not significant, so no ARCH model is actually returned. Allin.

Suppose I want a function that may be applied to one series (for example, "income") so that the function produces several output series (for example "trend", "seasonal", etc). The function may be: function decomp(series y, list comp) # some calculus ... series y_trend = ... series y_seasonal= ... list comp=y_trend y_seasonal return list comp end function at present, if you run this type of function on a series, for example: list components=decomp(income) you get a new list "components" with elements "y_trend" and "y_seasonal". The question is: Is there a way to pass the name of the variable to the function, so that the function can generate new series which change the names depending on what series are applied to? For example obtaining names "income_trend" ... when applied to "income" and names "consume_trend"... when applied to consume. -- Ignacio Diaz-Emparanza DEPARTAMENTO DE ECONOMÍA APLICADA III (ECONOMETRÍA Y ESTADÍSTICA) UPV/EHU Avda. Lehendakari Aguirre, 83 | 48015 BILBAO T.: +34 946013732 | F.: +34 946013754 www.et.bs.ehu.es

On Thursday 14 June 2007 17:37:32 Sven Schreiber wrote: Ok, but in that example "endogenous" is a list, so this implies that the user should define it first: list endogenous = income list components = decomp(endogenous) and I think that, given that my function does not need more than one input variable, gretl should have a way to obtain the same effect with only one line: list components = decomp(income) May be the function "varname()" of the "sprintf" command can help in doing something like that? -- Ignacio Diaz-Emparanza DEPARTAMENTO DE ECONOMÍA APLICADA III (ECONOMETRÍA Y ESTADÍSTICA) UPV/EHU Avda. Lehendakari Aguirre, 83 | 48015 BILBAO T.: +34 946013732 | F.: +34 946013754 www.et.bs.ehu.es

On Thu, 14 Jun 2007, Ignacio Diaz-Emparanza wrote: I may be missing something, but I think the answer is that you'll have to pass a list to achieve what you want here. The trouble is that when a single series is passed to a function, this series is known inside the function by the name of the corresponding function parameter, _not_ by the name it has in the calling script. function foo (series y) <create list of variables based on y> return list modlist end function list foolist = foo(myvar) In the above, function foo has no way of knowing that the series it sees as "y" is known as "myvar" in the caller. One way around this would be to allow string arguments to user-defined functions. Allin.

I finally made it to the computer where I'm running the regressions. When I run an OLS regression on the model I'm testing and run the arch test I get this output: Test for ARCH of order 24 PARAMETER ESTIMATE STDERROR T STAT P-VALUE alpha(0) 2.08514E+06 515100 4.048 0.00005 *** alpha(1) 0.938152 0.0247825 37.855 <0.00001 *** alpha(2) -0.0826077 0.0339689 -2.432 0.01513 ** alpha(3) -0.0205300 0.0340354 -0.603 0.54646 alpha(4) -0.0667144 0.0340145 -1.961 0.05001 * alpha(5) 0.0127966 0.0339762 0.377 0.70650 alpha(6) -0.0590282 0.0337789 -1.747 0.08074 * alpha(7) 0.0648604 0.0338031 1.919 0.05519 * alpha(8) 0.0209728 0.0338221 0.620 0.53528 alpha(9) -0.0761478 0.0337007 -2.260 0.02398 ** alpha(10) 0.0724101 0.0336583 2.151 0.03160 ** alpha(11) -0.0280495 0.0337054 -0.832 0.40542 alpha(12) 0.0116758 0.0337173 0.346 0.72917 alpha(13) -0.00739561 0.0337165 -0.219 0.82641 alpha(14) 0.0106318 0.0337551 0.315 0.75283 alpha(15) 0.103602 0.0337038 3.074 0.00215 *** alpha(16) -0.116552 0.0337181 -3.457 0.00056 *** alpha(17) 0.00503461 0.0338357 0.149 0.88173 alpha(18) -0.0265224 0.0337809 -0.785 0.43249 alpha(19) 0.141321 0.0337470 4.188 0.00003 *** alpha(20) -0.0856709 0.0339276 -2.525 0.01166 ** alpha(21) 0.0519347 0.0339545 1.530 0.12633 alpha(22) -0.0276832 0.0339709 -0.815 0.41524 alpha(23) 0.0396563 0.0339167 1.169 0.24248 alpha(24) -0.00949085 0.0246846 -0.384 0.70067 Null hypothesis: no ARCH effect is present Test statistic: TR^2 = 1182.98 with p-value = P(Chi-Square(24) > 1182.98) = 1.04224e-234 The p-value is essentially zero, so I should be rejecting the null hypothesis and there should be a model availalbe for forecasting when I run the arch command then correct? Thanks, Chris On 6/13/07, Chris <quilley(a)gmail.com&gt; wrote:

On Sat, 16 Jun 2007, Chris quoted me: Sorry, that "explanation" is wrong. The arch command _used to_ work that way, but I rationalized the command at some point and now it's more standard: "arch" gives you an ARCH model, whether or not the ARCH effect is significant. If you want to test for ARCH, you use the "lmtest" command with the --arch option. The reason you were getting the message "no model has been estimated" is that in rationalizing the arch command I didn't get all the details right. This is now fixed in CVS and the current Windows snapshot: http://ricardo.ecn.wfu.edu/pub/gretl/gretl_install.exe Allin

Thank you very much for fixing the bugs. My one concern about deprecating the arch command is that it estimates much more quickly than the garch command (although using garch 0 24; stuff should be the same as using arch 24 stuff, in practice it doesn't estimate at anywhere near the same speed and indeed the garch command above can fail when the arch command doesn't). I think this is because the garch command uses a complex iterative approach to estimation, where the arch command does something else, but I'm only vaguely aware of this time series stuff, I know just the barest amount I've been able to cram into my head over the last little while. Chris On 6/16/07, Riccardo (Jack) Lucchetti <r.lucchetti(a)univpm.it&gt; wrote: