Timeline for Does non-stationarity in logit/probit matter?
Current License: CC BY-SA 3.0
22 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Feb 21, 2020 at 2:45 | history | edited | kjetil b halvorsen♦ |
edited tags
|
|
| Jan 30, 2018 at 21:27 | answer | added | stratozyck | timeline score: 1 | |
| Feb 10, 2017 at 19:16 | answer | added | Dave Harris | timeline score: -1 | |
| Jan 31, 2017 at 19:31 | answer | added | Giuseppe | timeline score: 5 | |
| Apr 7, 2016 at 16:31 | answer | added | Cerbero | timeline score: 6 | |
| Apr 19, 2015 at 12:22 | comment | added | Jirka | Maybe I should have written that the key non-stacionary variable is share of foreign direct investment to total foreign liabilities. It is not stacionary, but it is always between 0 and 1. Does this makes some difference? Maybe non-stacionarity does not matter, if it is guaranteed that it can not rise to infinity (?) | |
| Apr 19, 2015 at 12:21 | comment | added | Jirka | Thanks, I will study some sources, you are recommending. I tried to perform linear probability model, and when I include time fixed effects, the key non-stacionary variable and one stacionary variable become insignificant. Including random time effects it is OK. Including time fixed effects + cross section fixed efficts the key variable is significant. So it is a bit confusing... | |
| Apr 19, 2015 at 0:44 | comment | added | Zachary Blumenfeld | Also it may be worth your time to do some background research for non-linear panel data models with possible non-stationary independent variables. I am pretty sure Wooldridge does research along these lines and Greene has some stuff out there concerning time controls for logit/probit. Newey is another big name in this stuff too. I am sure there are many others as well... | |
| Apr 19, 2015 at 0:39 | comment | added | Zachary Blumenfeld | When it comes to first differencing, this is sort of like "de-meaning" the time series variables which is very similar to the idea of adding fixed effects... controlling for changes in level for each individual time period. However, formal treatment of this sort can be quiet complex and there is still a chance for autocorrelation in residuals <a href="people.stern.nyu.edu/jsimonof/classes/2301/pdf/…> has some sensible suggestions using pearson residuals to check for autocorrelation. | |
| Apr 19, 2015 at 0:06 | comment | added | Zachary Blumenfeld | I am not 100% sure. The idea with the linear model is that you can easily do residual diagnostics for non-stationarity (or auto-correlation) in the residuals. The average marginal effects for the logit should be extremely close to the linear model margins. Using this fact, if you know the linear model is unbiased (via checking the residuals), than you can say the logit is estimating average marginal effects unbiasedly as well (given you observe the average margins are pretty much the for both models). Thus, it would seem reasonable to believe the logit is unbias. | |
| Apr 19, 2015 at 0:01 | history | tweeted | twitter.com/#!/StackStats/status/589579415564247040 | ||
| Apr 18, 2015 at 22:35 | comment | added | Jirka | So - if linear model is OK, or if I fix the problem with some way, do you think I can use the model without any other testing? (I tried a to avoid multikolinearity using correlation matrix, It is not enough, I know, but it should be OK) Thanks. | |
| Apr 18, 2015 at 22:35 | comment | added | Jirka | Thank you very much, I will try it. If the linear model will not be OK, maybe the easiest solution of non-stacionarity would be to use first differences. I have tried it now, and I felt relieved, that my key variables kept their significance... So if other solutions you are proposing appears being too difficult for me, this should fix the problem. But of course, I would prefer to keep those variables at levels, it has better economic sense as stock variables... | |
| Apr 18, 2015 at 21:48 | comment | added | Zachary Blumenfeld | Also there are some interesting things you can with re-sampling to reduce incidental bias in non-linear panel models which I think well apply to your problem. <a href="onlinelibrary.wiley.com/doi/10.1111/j.1468-0262.2004.00533.x/…> is a paper by Hahn and Newey that does so with a jacknife. I would not know how to implement something like this in Eveiws or STATA (Sorry), but I am sure it's do-able in STATA. | |
| Apr 18, 2015 at 21:40 | comment | added | Zachary Blumenfeld | They aren't allowing fixed effect in the logit because they are trying to avoid incidental parameter bias which on second thought makes good sense. One thing you could do is run a linear probability model, with and without random/fixed time effects. You can do residual diagnostics for these models and see if your margins of interest change. If your within group residuals are uncorrelated and time controls do not significantly change the margins than you may be fine with the logit (the average margins should for the logit should be similar to the linear model) | |
| Apr 18, 2015 at 20:46 | comment | added | Jirka | I know, how to test resisuals in time series models, like autocorelation, heteroscedasticity, normality, but I do not know what is necessary to do performing logit. Morover in papers similar to mine i have seen authors are testing nothing and they are using also for example loans/GDP, which are non-stacionary (at least in my dataset) I will be glad for any advice, thank you. | |
| Apr 18, 2015 at 20:42 | comment | added | Jirka | Thank you for your quick reply. The problem is, that Eviews do not allow any fixed/random effects performing logit/probit on panel data. I tried also to use Stata, it offers fixed effects, but i suppose it is not time effect, but cross section one. (Moreover it drops many countries which have not had any crisis, i.e. all dependent variables are 0) Without any selection stata uses random effects, but the outcome is nearly the same as Eviews outcome without any effect. The results I have ignoring non-stacionarity is nice,mostly according to theories... | |
| Apr 17, 2015 at 21:53 | comment | added | Zachary Blumenfeld | Is there any reason you cannot use time fixed effects? Doing so often controls for changes in level of non stationary variables. The logit uses maximum likelihood which assumes iid latent errors. If the errors are correlated through time it could result in a biased fit regardless of how standard errors are calculated. Again using time fixed effects should account for most of this. | |
| Apr 17, 2015 at 19:42 | history | edited | Jirka | CC BY-SA 3.0 |
added 21 characters in body
|
| Apr 17, 2015 at 19:02 | review | First posts | |||
| Apr 17, 2015 at 19:04 | |||||
| Apr 17, 2015 at 19:02 | history | edited | Sycorax♦ | CC BY-SA 3.0 |
spelling
|
| Apr 17, 2015 at 19:01 | history | asked | Jirka | CC BY-SA 3.0 |