Likelihood Ratio Test P Value

[mjones1985]

Hi! I have an important question about the p value for the likelihood ratio test.

The command is
scalar lr =-2*(RESTAOECD1.@logl - UNRESTAOECD.@logl)

the value is -1,403

to obtain the p-value for the lr test result i have used the formula

scalar lr_pval=1-@cchisq(-1.405855922849895,1)

It says that there shouldnt be negative numbers. How do i solve the scalar lr_pval formula then?


Another question: The likelihoodratio test nullhypothesis is that a restricted model equals an unrestricted model, right?

so if the p-value is high, like 0,4 or so on, it would mean, that the nullhypothesis cannot be rejected, and the models are equal?
Thanks for your help!!

[startz]

Something is probably wrong in your estimation. The restricted log likelihood can't be higher than the unrestricted, since the possible parameter values for the latter include the restricted estimates.

[mjones1985]

Ok thank you!

i want to model a stochastic trend like : μ_t=μ_(t-1)+ η_t mit η_t~ normalverteilt(0,σ_η^2)
i generated to series, one nrnd fpr η_t and one for μ_t = nrds
where is the mistake?

[startz]

Ok thank you!

i want to model a stochastic trend like : μ_t=μ_(t-1)+ η_t mit η_t~ normalverteilt(0,σ_η^2)
i generated to series, one nrnd fpr η_t and one for μ_t = nrds
where is the mistake?

You may want to show your actual code to see if someone can find a bug.

[mjones1985]

The unrestricted model ist LOGPECAPOECD = C(1)*STOCHASTICTRENDFINAL + C(2)*LOGGDPOECD + C(3)*LOGPMAXT + C(4)*LOGPRECT + C(5)*LOGPCUTTFINAL

where LOGPECAPOECD is the natural logarithm or per capita gdp, and stochastictrendfinal is the stochastic trend generated as: stochastictrendfinal = stochtrend(-1) + stochtrend2, in which stochtrend and stochtrend2 where generated nrnd series.

I would like to test, rather the stochastictrend is important for the model, or a constant is better.
But if i substitute the stochastic trend for a constant c, the model is not restricted, because the likelihood gets bigger as you pointed out.


another question i have, what can i do with the natural logarithm of a negative series?
i want to use the log of negative growth rates, my only possible way to use them in this model was to log the positiv numbers, and multiply them with -1, but i think thats crap :(.
But other authors used it before, can you imagine how they did it?