System Architecture

The prediction of a future time series relies on symbiotic algorithms that belong to either the core or the learning type. Learning algorithms control the model optimization process. Different stocks exhibit different chart patterns that can be mathematically defined. Learning algorithms explore sequential relationships of the patterns as well as permutable symmetries found within the time series data. Learning algorithms effectively create a digital signature which contains arrays of parameters that are unique for every stock or index.

Data processing stages that utilize core algorithms can be aggregated into three groups. Initial data mining is the most processor-intensive task since it involves creating symmetry channels. Each symmetry channel is bound by two parallel lines formed by consociating time series data swivel points. Once established, each channel is subject to iterative comparison processing for every high and low point in the time series – both within and outside of the bounds – in order to define outside and inside symmetry properties of the channel.

The next core algorithm processing stage involves extending the symmetry channels into the future and layering them so as to form a polynomial neural network at their intersections. Every neuron in the network has eight inputs and represents a single value point in future time, as defined by the intersecting symmetry channels. Core algorithms create neural connections by maintaining future symmetry properties of every channel, which are inherent in each neuron. The parameters of learning algorithms are factored in this processing stage, where lowest-density connections are replaced by the ones that have higher chain factor values.

In the last stage, core algorithms convert the neural network into a linked transition matrix that contains the most probable future time series data connections. Each future data point within the matrix contains link reference priority data which determines its time and value coordinates in the projected future time series and its corresponding connections to other future time series points.

The quantitative analysis is performed quadripartitely as it comprises monthly, weekly, daily and intraday stock market forecasting. Each given interval processing output data is applied in successive calculations as an input parameter.

Implementation

Once converted to a linked transition matrix, computational results are propagated over inteliCharts database servers. Validated terminal requests receive the data, update their own local SQL databases and synchronize them.

Since every point in the linked transition matrix contains reference information linking it to other points within the matrix, iteliCharts algorithms recalculate reference priorities based on the new transaction information received during market hours. Graphical User Interface interprets the data in the form of chart extensions, dynamically allocating and linking future time series data according to the transition matrix reference priorities.