Binary logistic regression - do i have to perform tests like in multiple regression? I am a beginner here. I would like to ask if i run binary logistic regression, do i have to do the tests below just like in multiple regression?
(1) stationarity of data
(2) multicollinearity
(3) autocorrelation
(4) heteroskedasticity
Thank you so much for your help!
 A: No.  Logistic regression makes no assumption about the probability distribution of the input features(variables), so you can have skewness, non-normality, etc. among your features.  Logistic regression is also not a linear method, and in fact is based on logarithms -- so it's multiplicative, i.e. based on log(x), exp(x).  Make sure there are not more features than records, i.e., $p \gg n$, or the program you're using will drop features due to singularity issues.  
A: Logistic multiple regression is conceptually similar to standard multiple regression, so you do have to evaluate many of the same quality-control metrics in either case, particularly with respect to the predictor variables. The difference is that in logistic regression you are trying to use a linear combination of the predictors to estimate the log-odds of the probability of an event, rather than an observed continuous response variable. As the observations are simply whether the event happened or not, you find values of the regression coefficients that maximize the likelihood of observing the yes/no outcome data through an iterative process. In standard multiple regression, the least-squares criterion gives effectively a one-step answer to the same optimization. The Brier score, often used to evaluate the goodness-of-fit of logistic regression, is effectively the mean-square error around the predictions. 
In an application to time-series data, however, logistic regression might not be the best choice for a yes/no outcome. Survival analysis, or the related recurrent-event analysis if the same event can happen to an individual at multiple times, may be much more useful as it also includes information about times to events and not just whether an event happened. In survival analysis there are standard ways to account for intra-individual correlations in event probabilities (clustering, frailty analysis) that would otherwise appear as auto-correlations. Depending on the structure of your data and your understanding of the subject matter, survival analysis could be preferable to logistic regression.
