TIME SERIES FORECASTING USING INDICATOR ARRAYS IN PYTHON

An analysis of methods of forecasting financial time series was carried out, models using intelligent technologies were considered. A proprietary model for forecasting time series using arrays of indicators in Python is proposed. Time series forecasting is the application of a model to predict future values based on previously observed values. And although regression analysis is often used to test the theory that the current value of one or more independent time series affects the current value of another time series, this time series analysis is not called "time series analysis" but focuses on comparing the values of one time series in different moments of time. The rate of the EUR/USD currency pair for 08/03/19 - 03/03/20 was taken as test data. After reviewing existing methods for time series forecasting using Python, the LSTM model was chosen. After analyzing the obtained results, the received signal forecast can be used as a decision support system. The accuracy of forecasting is primarily achieved due to the use of the LSTM model, normalization of the input dataset, the use of auxiliary technical indicators and the selection of parameters for the forecasting model. After reviewing existing methods for time series forecasting using Python, the LSTM model was chosen. Using LSTM, time series forecasting models can predict future values based on past sequential data. This provides greater accuracy for demand forecasters, leading to better business decisions. NVIDIA's cuDNN solves the computational problem. From experience, cuDNN can deliver 10+ times faster than standard use of CUDA directly. Based on the fact that the predicted Close price values are not accurate enough, signals were designed to recommend buy/sell positions. This signal represents a value between 0 and 1, where 0 is a buy and 1 is a sell.