Maximum number of iterations exceeded

[Kan David]

Hello,

I hope you guys are doing well.
I am estimating a model using panel ardl techniques, namely the dynamic fixed effect regressor. And Stata sent this message " Maximum number of iterations exceeded".
I would like to know what it means and how can I overcome this situation?

Thank you so much

[Clyde Schechter]

I'm not familiar with this particular command, but, in general, Stata estimation commands that require iterative calculations to arrive at a solution have some preset maximum number of iterations they will go through before giving up. Each command has its own default value for that maximum, set large enough that any calculation that fails to converge in that number of iterations is very unlikely to ever converge. So, unless this command has, for some odd reason, a small limit, you are probably looking at a model that simply cannot be estimated from your data. There are a few things to think about:

1. Look at the output that you got before that error message. Are there any coefficients or standard errors that are ridiculously large or ridiculously small? If so, that is usually a good sign that those variables are causing problems, and omitting them from the model and trying again may well lead to success. (You are then left with figuring out what is wrong with those variables that caused them to make the model inestimable. Often the data are riddled with errors, but sometimes there is no obvious reason.)

2. If that approach doesn't work, consider specifying starting values. An estimation can sometimes converge from a good enough starting point, but not from others. Some estimation commands allow you to specify your own starting values for the estimation in an option--check the PDF documentation to see if this command supports that. It will usually be listed under "estimation options" or "maximize options."

3. While you're looking at the estimation options, tinkering with some of those may also get you convergence where the default settings will not. This is worth at least a try.

4. If that doesn't work, you need to simplify the model. So start with a model containing only one independent variable, and then add new ones in, one at a time, and go as far as you can go with that and still get convergence.

5. This probably won't work, but, take a look at the iteration log near its end and see if the likelihood (or whatever is being revised in each iteration) is progressing from one iteration to the next. If so, you can try telling Stata not to quit so soon: this can be done with most estimation commands by specifying the -iterate()- option, and using a larger number for that than Stata give up at. (If the likelihood is not changing from one iteration to the next, or is going around in circles, this approach definitely will not be helpful. Frankly, it's unlikely to help either way, but in the other case there is at least a chance it will work.)

If none of this works out for you, you could try posting back showing the output Stata gave you (even though the estimation did not converge and the results are not valid, they sometimes contain clues to the source of the problem). Somebody might be able to spot a problem there (other than ridiculously large or small coefficients or standard errors).