Preprocessing of Input Data for the Stock Market Monitoring System

Research Problem Formulation. Modern financial markets are characterized by considerable complexity of the processes occurring in them. There is a globalization of international markets; currency, interest rate, rate of securities, staple prices volatility are increasing, and as a result, financial markets have become more volatile, complex and risky. An instrument of monitoring system of a stock market is needed in order to effectively control its state. One of the newest methods for studying time series is recurrence quantification analysis (RQA). Previous study showed the ability of measure laminarity (LAM) of RQA to reveal different periods of financial market functioning and to analyze crisis events on them [1, 2]. Financial market is highly susceptible to turbulence, and therefore the dynamics of the corresponding recurrence measures will contain the stochastic component. When building the monitoring system it is necessary to provide smoothing of LAM in order to automate the detection of critical points of transitions between periods of market functioning. Recent Research and Publications Analysis. Methods of processing and time series analysis are studied in the works of famous Ukrainian and foreign scholars and experts, including: John Yule, M. Kendall, A. Sewart, W. Hartley, J. Pollard, S. Ayvazian, G. Kildishev and others. However, the choice of strategies and methods for pre-processing and analysis of time series depends on the researcher’s ultimate goal. Purpose of this paper is to review, analyse and determine the optimal method of pre-processing (smoothing) data series for monitoring system of the stock markets. Justification of Scientific Results. For long series, it is usually impossible to specify an appropriate parametric curve to smooth the series for its entire length. In this case, one uses a variety of nonparametric methods for the analysis of time series, such as moving average smoothing, frequency filtering, etc. [3]. Let us analyse the most common methods of nonparametric smoothing of time series. Moving Average. While smoothing with this method the actual values of a dynamic series are replaced with mean values, that characterize the midpoint moving period [4]. Simple smoothing is based on the creation of a new series of simple arithmetic mean, calculated for time periods with the length k: