A Typology of Very Small Companies Using Self-Organizing Maps

Abstract

The world of very small companies (VSC) is characterized by a great diversity in terms of quality and behavior. Consequently, the principles for economic evaluation used today on medium to big sized companies cannot directly be used on VSC that must be grouped with companies of similar behavior. The purpose of the present study, conducted for the French Ministry of Finance, was to find an original technique to group these companies into clusters, with an economic theory background. Self Organizing Maps (SOM) were used to create subsets of companies from the original database. The method has been implemented in Java software with a web front-end to perform the computer evaluation of the technique. The method is validated on a data set of 459 companies collected by French students. Using the clustering software we found four subsets, that correspond to the economic theory of 'Worlds of production': interpersonal, industrial, immaterial and merchant world.

Keywords: Small companies, data analysis, neural networks

JEL Classification: D81, C45, C81, L2