RUS   UKR  
DonNTU   Masters' portal

Abstract

Content

Introduction

The development of financial markets holds a special place in the emerging market economy. Thus, the securities market allows governments and enterprises to significantly expand the range of sources of financing, not limited to budget funds, self–financing and Bank loans. If we consider the World experience, at the end of the twentieth century, the stock market became the main source of investment resources in dynamically developing countries. The securities market, allowing the transformation of savings into investments and the transfer of financial resources between sectors of the economy, contributes decisively to economic growth and welfare of the population.

1. Theme urgency

TIn modern conditions of formation of the Russian stock market closest to us, research on modeling the forecast of quotations of securities is of particular importance. Fluctuations in stock indices, the crisis of mortgage lending in the United States and other shocks to the securities market show that the need for these studies is ripe and relevant. Both in Russia and in the leading countries, fluctuations in this market are less dependent on political influence and the influence of other non-market factors, which confirms the need for objective research in this direction. Developments on this subject can be useful for both legal entities and specific citizens [1].

2. Neural network-based forecasting

In recent years, financial analysts began to be of great interest to so-called artificial neural network – mathematical models and their software or hardware implementation, based on the principle of organization and functioning of biological neural networks-networks of nervous cells of a living organism. This concept arose in the study of the processes occurring in the brain in thinking, and in an attempt to simulate these processes. Subsequently, these models were used for practical purposes, as a rule, in forecasting tasks. Neural networks are not programmed in the usual sense of the word, they are trained. The ability to train – is one of the main advantages of neural networks over traditional algorithms. Technically, the training is to find the coefficients of connections between neurons. In the learning process, the neural network is able to identify complex relationships between input and output data, as well as perform generalization. The ability of neural networks to the prediction of follow directly from its ability to synthesis and release of the hidden dependencies between the inputs and outputs. After training, the network is able to predict the future value of a certain sequence based on several previous values and/or what factors exist at the moment. It should be noted that forecasting is possible only when the previous changes are really to what extent predetermine the future. For example, prediction of stock prices based on price quotes over the past week may not be successful, while the prediction results of tomorrow's lottery, based on data over the last 50 years almost certainly will not give any results.

2.1 Deep learning

Deep learning is a set of machine learning algorithms that model high–level abstractions in data using architectures consisting of several nonlinear transformations. Deep learning technology is based on artificial neural networks (ANNs). These ins receive learning algorithms and ever-increasing amounts of data to improve the efficiency of learning processes. The larger the amount of data, the more efficient the process. The learning process is called deep, because over time the neural network covers an increasing number of levels. The deeper this network penetrates, the higher its performance [7].

Network trained using deep learning algorithms, not just superior in accuracy to the best alternative approaches, but also in a number of tasks showed the beginnings of understanding of the meaning of the information supplied (for example, for image recognition, the analysis of textual information and so on). The most successful modern industrial methods of computer vision and speech recognition are based on the use of deep networks, and the giants of IT–industry, such as Apple, Google, Facebook, buy teams of researchers engaged in deep neural networks.

2.1 A restricted Boltzmann machine

A restricted Boltzmann machine abbreviated RBM– a kind of generative stochastic neural network that determines the probability distribution on input data samples.

A restricted Boltzmann machine

figure 3 – an Example of a graphical representation of a Boltzmann machine. In this example, 3 hidden and 4 visible neurons [9].

A feature of restricted Boltzmann machines is the ability to be trained without a teacher, but in certain applications limited Boltzmann machines are trained with a teacher. The hidden layer of the machine is a deep feature in the data that is detected during training [10].

Conclusion

At this stage of the master's work was carried out analysis of the process of forecasting stock quotes in the conditions of the exchange.It was conducted a study of the exchange, on what principle it works, analyzed the data used in forecasting. The main algorithms that solved this problem were analyzed. The methods of forecasting with the help of neural networks are considered. The analysis revealed that neural networks with deep learning showed the best results.

In the future, the software implementation of the developed neural network and experimental finding of optimal parameters to achieve its maximum efficiency is expected.

References

  1. Сергиенко, А.Г. Моделирование прогноза значений котировок ценных бумаг с использованием нейроных сетей [Электронный ресурс] / А.Г. Сергиенко. – Электрон. текстовые дан. – Режим доступа: свободный; http://economy-lib.com/...
  2. Методы фундаментального анализа [Электронный ресурс] / Иркутский В.. – Электрон. текстовые дан. – Режим доступа: свободный; https://equity.today/...
  3. Методы технического анализа [Электронный ресурс] / FRANKIEKRAVETS. – Электрон. текстовые дан. – Режим доступа: свободный; https://equity.today/...
  4. Японские свечи как метод анализа цены акций [Электронный ресурс] /. – Электрон. текстовые дан. – Режим доступа: свободный; https://studopedia.net/...
  5. Свечной анализ для определения тренда [Электронный ресурс] /. – Электрон. текстовые дан. – Режим доступа: свободный; https://forex-broker.com.ua/...
  6. Прогнозирование фондового рынка с использованием нейронных сетей [Электронный ресурс] /. – Электрон. текстовые дан. – Режим доступа: свободный; https://habr.com/...
  7. Глубокое обучение: что это такое? [Электронный ресурс] / О. Масловская. – Электрон. текстовые дан. – Режим доступа: свободный; https://stfalcon.com/...
  8. Третье поколение нейросетей: Глубокие нейросети [Электронный ресурс] / VLADIMIR PERERVENKO. – Электрон. текстовые дан. – Режим доступа: свободный; https://www.mql5.com/...
  9. Машина Больцмана [Электронный ресурс] /. – Электрон. текстовые дан. – Режим доступа: свободный; https://ru.wikipedia.org/...
  10. Ограниченная машина Больцмана [Электронный ресурс] /. – Электрон. текстовые дан. – Режим доступа: свободный; http://www.wikiwand.com/...