What is calculated on your PC

We have solved the problem of generating a lot of forecasts distributed over different computers by letting each PC compute the forecast horizon of one stock price per task (one day, one week, one month, and longer periods of time are also planned for long term investors). In order forecast a stock trend over a period of one day, one week, or one month, at least three MoneyBee® computers will need to have submitted good forecasts for the same stock (or index or currency).

Your PC then actually trains one entire neural net for exactly one stock trend and one forecast period (for configuration, see Lower left-hand Honeycomb). If it was trained successfully and well enough it is capable of submitting forecasts - these are, however, not visible on your screensaver, since this only displays past data. The real forecasts are compiled in the BienenStock® database. That is for your own safety - all forecasts need to be tested by us before they can be published. Valuable, and tried and tested forecasts can then be found at uk.moneybee.net in the Member Area.

For the time being, you cannot influence the computing process - we distribute the tasks as suits MoneyBee® best by preparing data packages and tying these together. These are then downloaded by the users one at a time for those users picking up the next packet - therefore, the stock you receive has been randomly selected. Each task packet contains several tasks that are processed in sequence - during the ten hour computing time (one packet) you might therefore be watching the net training for several different stocks.

The stock your computer is currently processing is listed in the centre display Honeycomb of the screensaver below the diagram in the text (directly under "Forecast"). The neural net on your computer is "shown" past stock sequences. In the Upper left-hand Honeycomb you will see the period where this training data is from. A few days separate this training date and the download time - these are data the net has never seen and is required to estimate. That is why the period of time in the centre display Honeycomb is displayed in yellow (right-hand half of the curve). On the basis of these dates (and all following days), the forecast quality can be calculated in our BienenStock® database. Here is a demonstration of a forecast that was good during training, yet bad afterwards, as can happen.

In order to acquire complete tables, all tasks need to distributed to several computers, and only the best forecasts will be implemented. The system is therefore redundant. That does not mean that the other computer results are worthless. A bad net is also valuable to us since in the comparison with other nets we can find solutions for how to configure the nets and tasks so that they submit the best possible results. We can then control the number of hidden levels, the selection of stocks taken into account, etc. If all members submit better results on the basis of this knowledge, we will be able to offer more forecasts (further time horizons, different markets). No computing time is wasted!

In the optimisation of artificial intelligence, our team is breaking new ground: since MoneyBee® is a worldwide unique project, we cannot refer to other sources of knowledge. In this sense, you are participating in an exciting mass experiment in economic science - and are profiting from the results. Of course, we will be keeping you constantly informed of developments concerning MoneyBee® as a program and as an innovative forecast system at uk.moneybee.net.