Kanban Cell Neuron network Stock Trading System (KCNSTS)

摘要

A novel system for machine cognition predicts stock trading signals. The system uses the Kanban cell (KC), Kanban cell neurons (KCN), and Kanban cell neuron networks (KCNN), patent pending. The KC is an asynchronous AND-OR gate without feedback and is self-timing. Input data is processed until input is equal to output. For the KCN, input data is a four-valued logic (4VL) based on the 2-tuple as four logical values in the set of {"", 01, 10, 11} of four-valued bit code (4vbc) with "" equivalent to 00. The algorithm is a linear, multivariate, clustering formula which is not piecewise continuous. Multiple KCNs in the KCNN emulate the human neuron with nine logical inputs and one output. The KCNN model in parallel is scalable for large data sets. The model is adaptable as a forward-looking rules-engine as based on bivalent trial and error. The real-time algorithm is implemented by a look up table (LUT) to occupy 64 KB in software. In hardware access to a sparsely filled LUT is minimized with a 2-bit value per logical signal to occupy 194 KB. On a $40 device the LUT processes at 1.8 BB KCNs per second or about 1600 times faster than in software. KCNN is applied to analytics for time series of econometrics as the Kanban Cell Neuron Stock Trading System (KCNSTS). Virtual examples are given for the prediction of trading signals. For 129-trading days, 24 Asian electronic traded funds (ETF) produced an annualized 6% return on 70 no-charge trades. For 49-trading days, one OTC stock produced an annualized 67% return on 10 trades.