I am doing a bond rating change model and and have read about a multinomial logit model which is similar to a regular logit model but with more than 2 choices. I was thinking of running for my dependent variable 0-no chng 1- rating move up 2- rating move down and regress this against changes in fundamental ratios to see which ratio best forecasts rating movements. Can anyone explain to me the calculation process for this model? If you have any experience with this type of model, do you know how to do it with EViews. Thanks.

You cannot do multinomial logir under Eviews. Eviews is predominantly a time series econometrics package. Less suitable for multinomial and nested logit models. LIMDEP is one of the predominant packages for multinomial logit models. You can try SAS or SPSS as well. (but you have to check)I would not be surprised if most of your results are not statistically significant and/or have low coefficients (no economic significance). Also ratio relationsips are possibly non-linear and/or interactive (e.g. Gearing ratio may work in combination with cash flow coverage - Gearing x CFDC).What do you mean by the calculation process of this model ??? A word of caution with the multinomial, 3 levels is fine but if you have two many levels (anything beyond 5 - your cross partial terms gets a bit numerous and you start getting rubbish for your results which defies practical interpretability)