Probit model with fixed effects

[Hans Huber]

Hello there,
my name is Hans and I'm currently making a research about defaults on credit for businesses from different industries and sizes. My main independent variable is the monthly activity of the economic, so I'm trying to predict the amount of default depending on the monthly activity of the economy. My hypothesis is that a better economic cycle means less proportions of default. Following that I want to control the effects of the economic activity by industry and size of the firm but I don't know how to make it using probits models, because I've been told that there are no fixed effects using binary models. So I was thinking something like this: "probit Default Ec_Activity i.industry", with Default as a binary variable (1 if they did default, 0 if they didn't), Ec_Activity as the monthly growth of the economy in %. and i.industry variables for the string variables for every industry that I previously encoded. With that been said, I don't know if the margins commands will give me the marginal effects that the economic growth has on the probability of default for every single industry or not, because I was thinking that the model I proposed won't do that. So right now I'm a little bit scared about misinterpreting the results that I'm getting

The other options that I was thinking about was making this model for every single industry: probit Default Ec_Activity if industry == "name of the industry" and then margins, dydx(*), but I have many industries so I was wondering having a unique code that could give me the effects of every industry and in which I could graph all of them.

Hope my english is understandable and that someone could help me.

Thank you very much !

[Carlo Lazzaro]

Hans:
welcome to this forum.
Some comments about your query:
1) I’m not clear whether you're dealing with a cross-sectional or a panel dataset;
2) be as it may, if your regressand is categorical, the usual tools are -logit- or -logistic- if you have cross-sectional data,or -xtlogit- if you're dealing with a panel dataset;
3) a single regression is the way to go;
4) last but not least, please take a look at and act on the FAQ to post more effectively. Thanks.

[Carlo Lazzaro]

Hans:
as an aside, please note that the -fe- specification in -xtlogit,fe- implies conditional fixed effect.

[Hans Huber]

Hello Carlo,
Thank you very much for your help and answer. In my case, I've a cross sectional data. So which would be the code to regress default with monthly economic activity controlling by industry and computing the marginal effect of the activity for every industry ? something like: logit Default Ec_Activity i.industry ?

Thank you very much again for your help !

[Carlo Lazzaro]

Hans;
I would go:

logit Default Ec_Activity i.industry

After that, take a look at the several examples on logistic regression that are reported in -margins- entry, Stata .pdf manual.