Selective Pattern Matching Method for Time-series Forecasting

The selective pattern matching method for forecasting the increment signs of financial time series is proposed. This approach is based on indexing the time series to find similar sites in their dynamics based on the K-nearest neighbors method and selective grouping of these sites by the increment signs observed when completed. Similar sites are identified by calculating measures of similarity between the supporting and non-supporting stories of time series. Depending on the representation of time series, Hamming measure or Euclidian measure can be used for indexing. Before applying the method, it is recommended to carry out the procedure of pre-forecasting fractal time series analysis for determining the levels of persistence, antipersistence and randomness of time series, identifying the availability of memory and determining the medium length of quasicycles. The parameters, defined based on the pre-forecasting analysis are used in forecast generation by the described method. The proposed method can be used as part of information forecasting and decision support systems, including those used in the currency market to improve the accuracy of forecasting the increment signs of time series