The Moneybee® Forecasts

Once a net has been trained successfully, it submits its results to our central database, the BienenStock® . The results essentially only reflect threshold values and weightings. Since these are pure numbers, that are also compressed, the result file is not very large: neural networks hardly carry any ballast with them.

In the BienenStock® , we can quickly reconstruct and use the neural network on the basis of the result values - the actual forecast is even quicker: within milliseconds a stock price diagram of the current day passes through the network and submits the forecast for the future - in the same way the network has practiced it millions of times and has learnt from it. On the basis of the quality of the past few days, the quality of the network is tested again - if it is bad, it is filtered out. Only good nets are used for the forecast.

The quality of the nets is determined by comparing the values that actually occurred with those for the forecasts. What is calculated is how many forecasts (in percent) corresponded to the real values within a certain tolerance range (variation). The quality value in percent therefore states: "With a probability of XX%, this forecast will be accurate" - those are the percentages in the forecast table behind the actual forecast. The quality in the Upper left-hand Honeycomb of the screensaver is ascertained somewhat differently.

For each forecast period (one day, one week, 30 day forecast) other tolerance values have to be defined. In our explanation one can see how the quality of the forecasts is calculated individually.