Maximum Likelihood estimation in EViews

[gbengaibikunle]

Hi all,

Please find attached an untitled .jpeg image containing a maximum likelihood specification I have been struggling with estimating in Eviews. I have been receiving a constant error message: "Missing values in @LOGL series at current coefficients at observation 2" And once I populate the observation, it simply goes on to complain about the next observation point. In the vent that all observation points are population, it instead screams: "OVERFLOW". It's been very frustrating for the last few days. Can you help please?

Thank you.

gb

[EViews Gareth]

It is doubtful you'll get much help without showing what you've done so far.

[gbengaibikunle]

Hi Gareth,
I have written a number of lines, but this is the only one that has given me any measure of result so far, however, the results are spurious. They do not include standard errors:

@logl logl1
f1=@exp(-c(4))*((c(4))^b/@fact(b))*@exp(-c(4))*((c(4))^s/@fact(s))
f2=@exp(-c(4))*((c(4))^b/@fact(b))*@exp(-(c(3)+c(4)))*(((c(3)+c(4)))^s/@fact(s))
f3=@exp(-c(4))*((c(4))^s/@fact(s))*@exp(-(c(3)+c(4)))*(((c(3)+c(4)))^b/@fact(b))

logl1=(1-c(1))*f1+(c(1)*c(2))*f2+(c(1)*(1-c(2)))*f3
@param c(1) 0.1 c(2) 0.1 c(3) 0.1 c(4) 1

Thank you.

Kind regards,

[trubador]

It looks like you are trying to estimate a probability of informed trade (PIN) model. In practice, if you want to estimate your model with maximum likelihood, then you should transform the likelihood function to ease the computational burden. For an example of logarithmic transformation of the likelihood function, please see: David Easley, Soeren Hvidkjaer and Maureen O’Hara (2010). "Factoring Information into Returns". Journal of Financial and Quantitative Analysis, 45, pp 293-309

Please also note that, factorial representations of variables easily lead to numerical problems (e.g. Overflow). It would be wise to transform these variables beforehand (e.g. b = b/1000).

[gbengaibikunle]

Dear Trubador,

Thanks for your help. I did try the @logit function in Eviews earlier and also took a log of the likelihood function with no results. I will read the paper you recommended and see if anything in there helps.
Once again, many thanks.

Kind regards

[trubador]

Simply taking the logarithm of the function will not work. I was referring to monotonic transformation. You can find the transformed version of the likelihood function in Easley et.al. (2010).