Related to my earlier request, the following works (perhaps clumsily), so maybe it will be useful to somebody else also. The full illustrative script is

<hansl>

nulldata 50
series res = randgen(N,0,1) #the conditioning variable
scalar sv = 0.3 #remains fixed
scalar thu = 0.5 #remains fixed
matrix param = ones(1,1) #initialize the object in the function "udensity"

function scalar udensity (const scalar sv, const scalar thu, matrix param, series res)
    scalar k = $t2
    densu = -0.5*param[1,1]^2/sv^2 - res[k]*param[1,1]/sv^2 -param[1,1]/thu + log(1-exp(-param[1,1]/thu)) # this is the conditional density of "param"
    return densu
end function

series modeuvals = 0 # declare the series to hold the results from the consecutive BFGScmax in the loop
matrix bounds = {1,0,$huge} #argmax cannot be lower than zero

matrix modeu = ones(1,1) #initialize the object we are seeking to compute. "modeu" will take the place of "param" in the "udensity" function

loop i=1..$nobs   --quiet
    smpl 1 i
    matrix modeu[1,1] = 0.1 #intiial value for the unknown value (argmax) we are seeking
    densval = BFGScmax(&modeu, bounds, udensity(sv, thu, modeu, res))
    series modeuvals[i] = modeu[1,1]
    smpl full
    endloop

</hansl>

Alecos Papadopoulos PhD
Athens University of Economics and Business
web: alecospapadopoulos.wordpress.com/

-------- Forwarded Message --------

Subject:	Re: functions and loops
Date:	Sun, 6 Oct 2019 21:50:37 +0300
From:	Alecos Papadopoulos <papadopalex@aueb.gr>
To:	gretl-users@gretlml.univpm.it

Good afternoon.

I need to find the argmax (the mode) of a conditional density, for many different values of the conditioning variable. For a single value of the conditioning series variable "res", the following block appears to work (pre-accompanied by the generation of the various inputs needed)

<hansl>

scalar ressc = res[1]
    function scalar udensity (const scalar sv, const scalar thu, matrix param, scalar ressc)
    scalar densu = -0.5*param[1,1]^2/sv^2 - ressc*param[1,1]/sv^2 -param[1,1]/thu + log(1-exp(-param[1,1]/thu)) #this is the conditional density of the "param"
    return densu
end function

    matrix modeu[1,1] = 0.1 #intiial value for the unknown value (argmax) we are seeking
    densval = BFGScmax(&modeu, bounds, udensity(sv, thu, modeu, ressc))

</hansl>

The function above uses the value that the conditioning variable "res" has at observation #1, and then BFGScmax finds the argmax given this value of "res". If I then set scalar ressc = res[2], I can get what I need given the value that the conditioning variable takes in observation 2, etc.

I was unsuccessful to transform the above script into a repetitive one, starting with the fact that I cannot define a function inside a loop, as Gretl informs me. I tried various other ways to essentially loop over a command like BFGScmax that requires as input a function that must change at each instance of the loop, but all failed.

Is there a way to do this?

-- 
Alecos Papadopoulos PhD
Athens University of Economics and Business
web: alecospapadopoulos.wordpress.com/