EViews.com

hi everyone,

I am currently making a model to find the probability of default of a firm. To do so I chose a logit regression (binary model) and find a final model keeping onlu the significative variables (Wald test used here). I have two doubts about my result if anyone could help me answering :

- if the p-value relative to the constant indicates that it is not significative, should I put the constant off from the model?
- I run the goodness-of-fit tests to see whether my model seems good or not, and i get

Prob.Chi Sq(8) = 0.3134 for HL- Statistic that indicates my model is good and
Prob.Chi Sq(10)=0.000 for Andrews statistic that is supposed to indicate the contrary.

Should I rejet then the model ?

Thanks a lot !

Eleo:
two remarks about your query:
- I would keep the constant in the model. The fact the it is not significant is not self-explaining in itself, as it may depend on what does it refer to in your model. Moreover, I would test if the constant is still unsignificant after centering the predictors around some relevant values of theirs (e.g. mean).
- your goodness-of-fit (gof) p-values shoud be considered in the light of the data generating process of your research topic. Hence, I woud take a look at which predictors were included in the righ hand side of the logistic regression in previous analogue research experiences instead of relying on gof test results only.

Kind regards,
Carlo