Multiple line-segments regression for stock prices and long-range forecasting system by neural network

This note proposes a forecasting system for stock prices by neural network with reference to multiple line-segments regression after examining simulation results of our modified forecasting system. Our objective is to construct middle and long time range forecasting system, we consider the week summary data instead of the ending-value of one day. Thus we modify our forecasting system replacing the ending-values of stock prices per day by the week summary data with the week average which are inputs of our neural network. Usual back-propagation method is used for the learning. After the learning, the associative memory of our neural network gives forecasting values for stock prices with their time unit to be one week. Unfortunately, this modified system does not yield satisfactory forecasting values in our simulation study. Hence, instead of forecasting stock prices as an output of our neural network, we set the tangent and the length of the multiple line-segments regression as outputs of our new neural network. Usual back-propagation method is used for the learning. After the learning, the associative memory of our new neural network gives forecasting values for stock prices by the collection of tangents and lengths of the multiple line-segment regression. Several simulation results are also discussed