Building iterative Neural Network forecasts
Artificial Neural Networks (ANN) models provide a robust method of generating forecasts. ANN can be built using nearly any input types including categorical, flag, and continuous inputs. ANN models are relatively insensitive to outliers and are capable of capturing subtle interactions between input variables. All of these benefits have made ANN models increasingly popular for many applications such as forecasting product sales, energy demand, spot market prices, and so on.
Even though ANN forecast models are generally superior to traditional forecasting techniques such as ARIMA, they do have a few drawbacks. The first drawback of ANN forecast models is that they are not autoregressive (as compared to ARIMA). The model builder must choose the appropriate lags for the input variables. For example, do we look at the price one day ago, one week ago, or one month ago when predicting the current price? The second drawback is that the ANN models predict...
