| Abstract [eng] |
In this paper dynamic time warping algorithm is discussed and its application opportunities for financial time series. The overview of published researches has shown that this algorithm in financial studies is not widely applicable. In this world the financial markets are becoming increasingly difficult to analyze using traditional methods, so there is a need of developing new techniques or using methods from the other areas that would let to investigate the financial time series and recognize their certain shape patterns. Therefore, in this work we investigate the application of dynamic time warping algorithm and its added value to analyse financial time series, especially when they are characterized in different speed or different time scales. In order to ensure the correct operation of the algorithm it was needed to verify if established path satisfied the main three criteria, also properly select local or global constraints depending from the data structure. After the three cases of experiments, we observed that the ordinary assymmetric and symmetrical restrictions are characterized by one point to one mapping, selection of a slope coefficient gave us one point to many points mapping. In the one practical application we found similar form of OMX return index behavior in the historical 2010 – 2013 year period. Also in this study case, 5 similar patterns of historical times series data were clustered and divided into 3 main groups according to their inherent time series pattern. Multidimensional time series case, where OMX price indices in the period of 2014 – 2015 years have been analysed, has shown that OMX Baltic index time series is most similar to the price index of OMX Ryga time series. According to the final work of the theoretical analysis and the research results, conclusions and discussions are provided at the end of this work. |
