Dynamic Cubic Neural Network with Demand Momentum for New Product Sales Forecasting

The various statistical models were developed for successful forecasts using historical demand data. However, if there are no sufficient or available historical demand data, for example brand new product, not many forecasting methods are available. For dynamic forecasting of the new product sales without the historical data, this paper proposes dynamic cubic neural network that consists of iterative modification mechanism for activation function and cubic architecture based on the concept of Bass model which describes diffusion processes of the new product with innovative and imitative buying behavior. In our model, the output scope of the activation function of hidden layer is modified for every period, according to demand momentum which is defined by demand inertia and price acceleration plays a key role in adjustment of output in iterative learning processes.