Abstract
This study uses artificial neural networks (ANNs) to reproduce aggregate per-capita consumption patterns for the New Zealand economy. Results suggest that non-linear ANNs can outperform a linear econometric model at out-of-sample forecasting. The best ANN at matching in-sample data, however, is rarely the best predictor. To improve the accuracy of ANNs using only in-sample information, methods for combining heterogeneous ANN forecasts are explored. The frequency that an individual ANN is a top performer during in-sample training plays a beneficial role in consistently producing accurate out-of-sample patterns. Possible avenues for incorporating ANN structures into social simulation models of consumption are discussed.