Endless Iteration problem (while I estimating pmg before Hausman test)

[MTaylan YAVUZER]

I may have given incomplete information as this is my first post on the forum. Sorry for this situation.

To do panel ARDL analysis, I have to choose between mg and pmg estimators. Therefore, before the hausman test, I wanted to predict with the code "xtpmg d.LGDP3 d.LGExp1 d.sdlgexp3 d.LPop1, lr(l.LGDP3 LGExp1 sdlgexp3 LPop1) ec(ECT1) pmg". I have 6 models in total. I ran the hausman test on my 4 models without any problems. However, in the remaining 2 models, it does infinite iterations. I had to break after 1000 iterations. N:7 and T:18 (annual), this number is too high.
I'm waiting for your recommendations. Thanks

. xtpmg d.LGDP3 d.LGExp1 d.sdlgexp3 d.LPop1, lr(l.LGDP3 LGExp1 sdlgexp3 LPop1) ec(ECT1) replace pmg

Iteration 0: log likelihood = 525.22957 (not concave)
Iteration 1: log likelihood = 531.3161
Iteration 2: log likelihood = 533.65613
Iteration 3: log likelihood = 536.58289 (not concave)
Iteration 4: log likelihood = 536.69607 (not concave)
Iteration 5: log likelihood = 536.7115 (not concave)
Iteration 6: log likelihood = 536.71345 (not concave)
Iteration 7: log likelihood = 536.71358 (not concave)
Iteration 8: log likelihood = 536.7136 (not concave)
Iteration 9: log likelihood = 536.7136 (not concave)
Iteration 10: log likelihood = 536.7136 (not concave)
Iteration 11: log likelihood = 536.7136 (not concave)
Iteration 12: log likelihood = 536.7136 (not concave)
Iteration 13: log likelihood = 536.7136 (not concave)
Iteration 14: log likelihood = 536.7136 (not concave)
Iteration 15: log likelihood = 536.7136 (not concave)
Iteration 16: log likelihood = 536.7136 (not concave)
Iteration 17: log likelihood = 536.7136 (not concave)
Iteration 18: log likelihood = 536.7136 (not concave)
Iteration 19: log likelihood = 536.7136 (not concave)
Iteration 20: log likelihood = 536.7136 (not concave)
Iteration 21: log likelihood = 536.7136 (not concave)
Iteration 22: log likelihood = 536.7136 (not concave)
Iteration 23: log likelihood = 536.7136 (not concave)
Iteration 24: log likelihood = 536.7136 (not concave)
Iteration 25: log likelihood = 536.7136 (not concave)
Iteration 26: log likelihood = 536.7136 (not concave)
Iteration 27: log likelihood = 536.7136 (not concave)
Iteration 28: log likelihood = 536.7136 (not concave)
Iteration 29: log likelihood = 536.7136 (not concave)
Iteration 30: log likelihood = 536.7136 (not concave)
Iteration 31: log likelihood = 536.7136 (not concave)
Iteration 32: log likelihood = 536.7136 (not concave)
Iteration 33: log likelihood = 536.7136 (not concave)
Iteration 34: log likelihood = 536.7136 (not concave)
Iteration 35: log likelihood = 536.7136 (not concave)
Iteration 36: log likelihood = 536.7136 (not concave)
Iteration 37: log likelihood = 536.7136 (not concave)
Iteration 38: log likelihood = 536.7136 (not concave)
Iteration 39: log likelihood = 536.7136 (not concave)
Iteration 40: log likelihood = 536.7136 (not concave)
Iteration 41: log likelihood = 536.7136 (not concave)
Iteration 42: log likelihood = 536.7136 (not concave)
Iteration 43: log likelihood = 536.7136 (not concave)
Break--
r(1);

[Clara Joan Joachim]

I am having the same problem, please advise if you found the solution. thanks

[Nick Cox]

Clara Joan Joachim

If model estimation doesn't converge, there is no universal or immediate fix. It may be that the model doesn't suit the data or the data don't suit the model. It may be that you need to try a simpler model or tweak the estimation procedure. There can't be more precise or useful advice without any details on data or model or code. Try a really simple model first. Tell us more about the data. I just answer here generally, and I am no economist or econometrician to give more specific advice on this kind of model, but you may get better answers from economists with more information.

[Clara Joan Joachim]

Nick Cox Thank you for your respond. I have a panel data of 8 countries for 30 years, with 13 variables. They are integrated at level and first difference. I wanted to try them with command xtpmg for panel ARDL model. Considering I need to examine which model is suitable (MG, PMG and FDE). I ran xtpmg on all of it before hausman test. However, MG and FDE showed max number of iteration exceeded. Please advice, thanksThank you

[Nick Cox]

Thanks for adding some detail. As said, this information is for economists to comment.

[Meenakshi Gautam]

hii Clara , am facing the same problem did you find the solution to your problem , kindly reply

[Nick Cox]

Clara Joan Joachim hasn't been seen since last July. She may reply now that she's been pinged. Otherwise #3, although not the answer you seek, is possibly as much as anyone can say.