NeuroStrategy FAQ
This site is best viewed in a browser that conforms to web standards.
Currently available applications for the analysis of financial instruments based on neural networks are not easy to operate. The user must define and prepare the input data and is usually also required to formulate and test the reaction. Frequently, the user must even determine the type and structure of the network and fine-tune dozens of parameters for training. These tasks require expert knowledge in areas such as statistics, mathematics, chart and financial analysis as well as experience in dealing with systems of artificial intelligence. NeuroStrategy relieves you of these complex tasks.
Building on our experience in time series analysis, for example, we have used a large number of data transformation processes for the preparation of the historical data of a security in the form that a mesh is able to evaluate. Now, as a user, you no longer have to deal with the problems of the calculation and testing of the so-called "market indicators". The selection of suitable formats of the input data as well as their preparation for neural processing are also done for you.
Our goal was to provide a universal system that is capable of handling as many securities as possible in a profitable way. In Version 1.1, NeuroStrategy now offers you three different types of meshes, each with an optional extension (component extraction).
The types
differ only in terms of the inputs used. While meshes of the UMI type use some inputs from the area of technical trading, meshes of the NMI type are supplied exclusively with data extracted through other preprocessing methods. The TSM meshes finally use inputs calculated from the original time series by means of time series modelling. Our extensive tests have shown that even the most commonly used indicators of technical trading are not useful with respect to many securities. There are, however, just as many securities, where such market indicators prove very useful for promoting the formation of a profitable strategy. We therefore recommend the use of all three available types in developing strategy models for a particular financial instrument and to compare these models and make a selection.
In addition to the basic types, NeuroStrategy also provides extended mesh types. These have a significantly higher processing capacity. Stated in somewhat technical terms, these mesh types are capable of extracting from the multidimensional space of the input data those dimensions (components) that contain the largest proportion of usable information. This allows useful conclusions to be drawn even from data with a paucity of informational content.
The mesh types with component extraction thus implement a type of filtering function with an additional preprocessing of the input data before this data arrives at the actual processing part of the mesh. These types are suitable for the analysis of securities, for which normal mesh types are unable to obtain a satisfactory result. Then and only then should they be used. For their use also harbors potential problems. Important information can be lost through the filtering, for example. Moreover, the significantly higher processing capacity can lead to an overspecialization (overfitting) on the training data. And finally, the training period of these mesh types is longer, since their neural network is substantially larger than that of standard types.
