Hi! I want to use logit regression whit lags of the independent variables. Now, I was told I need to use time fixed effects for this (even though I don´t get why). Is this true, and if so, how do I do it? I can't both set the data to time-series and panel (tsset and xtset), but have to choose. Also, can I go around this problem by constructing the lagged independent variables ouside of Stata, in Excel, and then xtset the dataset? I'm a novice at both econometrics and at Stata, so please explain to me as if I was a child?
Your dependent variable is 0-1. But note that you have a time series dimension, i.e. for the dependent a series of 0's and 1s over time. Note that a panel has a time series dimension in any case. Therefore, you could include fixed time effects, although they might not be necessary. In my view, it is better to explain the behaviour over time by some economic variables and not time effects, which are finally unexplained. You should be able to do all in Stata, including the specification of lags of the regressors. But of course, you can also prepare the input for the program in Excel.
This can be with generalized autoregressive score (GAS) models. Go to www.gasmodels.com for the papers explaining them. Key authors are Siem Jan Koopman, Drew Creel, Andre Lucas, Andrew C. Harvey, and T. Chakravti, as well as Francisco Blasques.
I am not sure if you have to use time fixed effect model. See .....whether one should use fixed effect or random effect model depends on some conditions. You should conduct “Hausman” test to decide between these two. Moreover, fixed effects itself can be of two types (1) time fixed effect [i.e value of parameter (in most cases it is intercept of ur model) changes across spaces but then those changed values remain same over time]. (2) Cross section fixed effect [i.e value of parameter changes overt time but then those changed values remain same across spaces (states, countries or district etc)]. Now the choice between time fixed effect and cross section fixed effect is not something that one can decide in priori. You have to conduct a test called likelihood ratio test to determine which one is appropriate for you. You may use Eviews 8.1 to handle these issues with ease.
Hello, are you talking about binary logit or multinomial logit?
Concerning time effects, I agree with Christian it is better in general to consider explanatory variables varying over time rather than time fixed effects, but it may (or may not) be the case that such observed variables are not sufficient to take into account time effects affecting simulatenously all your observations (I guess this is the reason why you were told to use time fixed effects).
What is the structure of your data? How many individuals? How many dates for each individual?
Dear Prof. Nordfors,
If your objective of doing Logit regression is to estimate the growth effects in the data then you can carry out both time series and time series analysis using panel data in an algorithmic (genetic) estimation procedure using the joint distribution of the error matrix using normal matrix inversion procedures. Please refer to my coauthored papers on mathematical statistical pricing and banking sector pricing on my RG page if you will. This will also have Econophysics procedures built in them and follow the Gauss Markov Theorem.SKM QC
Sir Soumitra Kumar Mallick will you please share your work realted to above answer as reference. I will looking out this kind of resources for my research article.
Thanks in Advance.
- Badges
- Science topic