A regression clustering method for the prediction of the pro capita disposal income in municipalities

The aim of regression clustering (Bin Zhang, 2003) is segmenting a number of units in some clusters in order to detect a good regression model in each cluster. Then regression clustering is suitable when, given some explicative variables (regressors), a single regression model doesn’t fit well all the units, but different regression models might fit well partitions of the data (see also Sarstedt and Schwaiger (2006)). In this paper a regression clustering procedure is adapted to a particular regression to predict the pro capita disposal income (PCDI) in municipalities. The particularity of this regression consist in: it is a two-level regression (municipalities and provinces) and the parameters estimation is run at the provincial level under some assumptions.