Red neuronal para clasificación de riesgo en cooperativas de ahorro y crédito

Abstract

In Ecuador exists a great number of credit unions (COAC) specifically 852, which are divided into 5 segments depending on their amount of assets. Nowadays, 66% of the microcredit obtained within the country corresponds to the cooperative system. Nevertheless, just 35 of 58 COAC (segments 1 and 2) present risk rating histories. The purpose of this research is create a neural network that achieves an acceptable percentage of accuracy, to classify a COAC within a scale of risk based on the value of its financial indexes; in order to helping the early detection of future problems. The artificial neural network (ANN) was fitted from data obtained through the Public and solidary economy Superintendence for COAC that presented a high index of assets. In addition, the history of quarterly risk ratings generated by rating agencies in the same period was used: January 2015 - September 2017. An ANN with a classification accuracy of 79.59% was obtained, percentage that is within the range of precision obtained by studies reviewed for classification activities in financial entities. The classification results could be further improved with the use of a hierarchical classification structure.