Title:
FORECASTING FINANCIAL TIME SERIES DATA USING MOTIF DISCOVERY AND MOTIF SEARCH

Authors:
Richard Britto, Rado Kotorov, USA

Abstract:

There are many methods, algorithms, and tools out in the market to predict financial data. While the main objective is to predict multiple time-steps into the future, it is hard to generate these predictions. The problem arises in multi-step financial time series prediction. Most techniques have been developed for a single-step prediction which is practical enough for decision making. This paper proposes a method of predicting prices multiple steps into the future using motif discovery and motif search techniques with the use of the Pearson Similarity algorithm to extract motifs and leverage the motif patterns to generate multi-step predictions. The experiment's results are compared to the Convolutional Long Short-Term Memory (ConvLSTM) model. The resulting proposed model appears computationally efficient and returned better prediction scores.