A METHODOLOGY FOR BUILDING NEURAL NETWORKS FOR STOCK MARKET PREDICTIONS

The article discusses the tools for constructing forecast models for stock markets. The prediction tool is neural networks. An assessment of the features of the use of neural networks for predicting financial time series is given. The methodol‑ ogy for constructing a neural network for stock market forecasting includes the processes of choosing the network architecture taking into account the significant factors of the system under study, the relationship of learning methods with the network topology, the formation of the training sample and determining the initial learning conditions. These factors together determine the quality of training of the neural network and the predictive properties of the model. Research into the practice of using neural networks in financial forecasts revealed changes in approaches to setting up a neural network: the transition from substantiating the choice of the type of network architecture to assessing the existence of a relationship between network settings and properties of the object of study; the use of adaptive algorithms for learning neural network. This allows you to make the neural model more flexible in the learning process and better reflect the relationship of the system under study. Promising areas for studying neural networks for forecasting financial markets are a combination of research methodology, financial time series, economic decision-­making processes, together with an analysis of the capabilities of neural models to reflect the totality of market factors due to the selection of network architecture, learning algorithms and evaluating the effec‑ tiveness of the model, Formation of the training sample, proper data quality and presentation methods are given.