The Simulation of US Consumer Credit Fluctuation Using Artificial Neural Networks

Abstract

The present paper shows the discussion and results of the research that simulated the fluctuation of the US Consumer Credit (CONS) using Artificial Neural Network (ANN). The research had several objectives, like: building, training and using an ANN as a possible tool for decision making, through the simulation of the US Consumer Credit. The condition for a successful training of the ANN was established as a smaller difference than 1.5% between the real data and the simulated data. A feed forward artificial neural network and a back propagation algorithm were used for the training and preparation of future use of the ANN. For the training result, two testing sessions were used. For the use of ANN in CONS forecasting, the research was extended with the simulation of CONS trend using trained ANN and a new set of consecutive values for each of the input data. Also, the new simulations determined a hierarchy of the inputs that were considered for the simulations of the CONS. In the conclusion, the researchers consider the ANN training and testing a success due to the values obtained: a difference of [-0.69; 0.32] % between the real and simulated CONS values. The trend simulation also shows the training success with accuracy smaller than 1.5%. The authors consider that the research can be extended to other countries or by adding others indicators.