Neural networks have been studied since the 1960s and are still a rapidly advancing field of study. The use of a neural network is an important non-linear method for projecting future values based on historical data (non-linear meaning "not using conventional linear statistics").

In a demonstration application a 100 day history of both the JSE and the NYSE was used to predict the price of the JSE for the following three days.

We have developed a precision tool-kit for constructing a wide array of neural networks that can be applied to forecasting problems in various applications.

Potential applications for forecasting are countless, however there is always the requirement that enough of the right kind of training data be available. This would include historical values of the series to be projected, as well as any other data series that may have had significant impact on its values.

A company with a complete sales history would be in a position to analyse both short term cyclical trends and long term market trends, thereby forecasting sales for the next period based on past performance.