Thread Subject: Distribution parameter estimation methods

Hi,

I just wondered what's the difference if for parameter estimation of a given time series and a supposed inherent distribution use either the 'mle' or the respective '...fit' method. I've tried both, and though the results are similar, there are slight differences.

Where do this differences come from and what would be the prefered choice?

Thank you.

Regards

Thomas

> I just wondered what's the difference if for parameter estimation of a
> given time series and a supposed inherent distribution use either the
> 'mle' or the respective '...fit' method. I've tried both, and though the
> results are similar, there are slight differences.
>
> Where do this differences come from and what would be the prefered choice?

The only difference that comes to mind is that the normfit and lognfit
produce a sigma estimate that is the square root of the unbiased variance
estimate. The mle function does maximum likelihood. So the former has a
divisor of (n-1), compared with (n) for the latter.

Did you notice differences for other distributions?

-- Tom

> The only difference that comes to mind is that the normfit and lognfit
> produce a sigma estimate that is the square root of the unbiased variance
> estimate. The mle function does maximum likelihood. So the former has a
> divisor of (n-1), compared with (n) for the latter.
>
> Did you notice differences for other distributions?
>
> -- Tom
>

I haven't tried yet, but I think you may have pointed out the answer, since the difference was very slight, though it was still there. I just was not sure how normfit worked.

Thanks & Regards

Thomas