Models of forecasting value of financial assets by using artificial neural networks

<p>The theme of the research is extremely important in the modern financial world, because prediction of time series is the basis for activities of many economic agents. The idea of investment assumes that while investing a sum of money in the current period we expect to get a future profit. Consequently, it implies some forecasting of the economic situation and future revenue assessment.</p><p>The current paper aims to develop the optimal algorithm on the basis of artificial neural networks (ANN or multilayer perceptrone) for predicting the sign of the yield for one step ahead, which will demonstrate reliable average efficiency.</p><p>The following purposes were put ahead:</p><p>o&nbsp;&nbsp; Learning the mathematical mechanism if the artificial neural networks</p><p>o&nbsp;&nbsp; Choosing the most suitable methods of such a forecasting algorithm validation</p><p>o&nbsp;&nbsp; Investigation of various forecasting algorithms based on multilayer perceptrone using real quotes of large Russian companies</p><p>o&nbsp;&nbsp; Drawing up an optimal algorithm and validate it on real data</p><p>The current research employs methods of numerical experiment using the language for statistical computing GNU R. As a data sources were used the daily quotes of large Russian companies from oil and gas sector: &laquo;Lukoil &raquo;, &laquo;Novatek&raquo;, &laquo;Surgutneftegas&raquo; и &laquo;Rosneft&raquo;. Data was extracted from the website finam.ru.</p><p>The artificial neural networks with different parameters were considered and also variants of their usage for predicting financial time series were examined. In the current paper the way of improving the forecasting efficiency was proposed.</p><p>As a result the predicting mechanism that predicts right yield sign in most cases was elaborated. The efficiency was confirmed be simulating the trading process and actions of some trader who is guided by the developed scheme.</p><p>The maximum value of the rate of return is 32% for period about a half of year.</p><p>This research confirms the prospects of applying ANN-based algorithms to forecasting tasks. The developed procedure can be used for practical purposes for decision-making on financial markets.</p>