Dean Wilkie posted on Sunday, September 10, 2017 - 9:18 pm
Excuse my ignorance but I am needing some guidance as to the most appropriate model for my research aim.
I have a data set consisting of 100 firms with their performance monitored weekly over 52 periods. I have estimated time varying coefficients for key variables for each firm and I am certain that there will be heterogeneity in the coefficients. Both between firms and over time periods.
Given that my aim is to identify the subgroups of firms with this data, what type of model do you recommend I use?
See especially the cross-classified analysis discussion during Day 1's part 4 and the clustering discussion during Day 2's part 7.
Dean Wilkie posted on Monday, September 11, 2017 - 9:19 pm
Thank you for your prompt response. I watched the presentation by Marten and I saw the similarities with my aim and his with the study on electricity consumption (i.e., clustering based on the values of the random coefficients). However, when discussing the clustering, he mentioned that he considered a number of techniques. Could you tell me what techniques he considered, or what you would recommend?
It depends on the objective of the clustering. I just view the estimates of the random coefficients as an efficient way of characterising the pretreatment data. I recommend that you consult the clustering literature to achieve your specific clustering objectives. In the particular small sample electricity example I mentioned, I tried e.g. kmeans and hierarchal clustering and simply looked at whether or not the clusters made sense from a consumption behaviour point of view, in discussion with my colleges.