0

I need some example on how to fill missing data with EM algorithm. The data, as the daily relative change of a stock price, assumed to normally distributed and univariate sample. I've done some literature search but hardly found any example on this. It seems when people talk about the application of EM algorithm to imputation of missing data, they usually give examples on multivariate case. These are the cases I see from most papers/lecture notes.

Now I am wondering whether people fill missing data for an univariate sample with EM algorithm, and whether the EM algorithm imputation is equivalent to mean imputation in this case. I really appreciate if you can share some insights or give a link for any reference on this topic.

TylerH
  • 20,799
  • 66
  • 75
  • 101
VincentHall
  • 37
  • 1
  • 1
  • 6
  • the MLE estimator for the univariate normal case is the mean. – alexwhitworth Oct 20 '16 at 04:10
  • 1
    When there is only one variable, then using EM does not really add anything to just analysing the available data. Univariate imputation (i.e., without any background information) is really just imputing "noise" based on the properties of the available data. That being said, EM is **not** equivalent with mean imputation! Mean imputation will actually make things worse because it will inflate sample size, leading to inferences that are too liberal. – SimonG Oct 21 '16 at 16:21

0 Answers0