The neural network is the technology that reproduces the human brain artificially. Time-series such as sound are widely studied using this. The neural network with the internal memory is suitable for time-series learning.
In this research, the neural network with the internal memory was made to learn the fluctuation in exchange rate as time series, and forecast that of the future.
As the result of the learning, the error of the network output didn't become small enough. So, the network that had learned to some extent forecast the fluctuation in exchange rate. The network couldn't forecast high and low successfully when they were used as supervisory signals.
Then, the network was made to learn high and the 21-day moving average line as supervisory signals, and forecast that of the future. As the result, it couldn't forecast the details, however, it could show the trend of the fluctuation.
Our future tasks are to improve the learning method, to input other information as supervisory signals and so on in order to make the network forecast the details.