Estimation of Regional Economic Convergence Equations Using Artificial Neural Networks with Cross Section Data

Theoretical developments and discussions on growth and regional convergence have been accompanied by another debate, associated with the type of data used and quantitative approaches adopted in empirical research. Estimation of convergence equations continues to play a key role in the study of economic convergence, despite criticisms. This paper introduces the use of artificial neural networks (ANN) in the study of convergence. It focuses on the concept of b-convergence and accepts that cross section data can provide useful information for its investigation. Non-linearities of the underlying relationships, the restrictiveness of assumptions on functional forms, and econometric problems in the estimation and application of certain theoretical models, advocate for the use of ANN algorithms. A back-propagation (BPN) artificial neural network is constructed and utilized to study convergence of regional, gross domestic products per capita in Greece, together with the application of a traditional econometric analysis. Cross-section statistical data on Greek prefectures are used while results and repeated testing show that the neural network performs very well in estimating convergence equations. It improves substantially the accuracy of estimates and predictability of the estimated relationships. In addition, the BPN algorithm could be used with time series or panel data, and it could estimate also convergence equations of additional economic or social variables.