Abstract

In stock markets, many types of time series models such as statistical time series model, fuzzy time series model, and advanced time series model based on artificial intelligence algorithms were advanced by academic researchers to forecast stock price. Some drawbacks are issued for these models as follows: (1) mathematical assumptions are required for statistical time series models; (2) the forecast from fuzzy time series model is a linguistic value that is not as accurate as statistical time series; and (3) a proper threshold is not easy to be produced by advanced time series model and the forecasting algorithm is unintelligible. To deal with these problems, we propose a novel price-pattern detection method to look for certain price-patterns (“price trend” and “price variation”) contained in time series variables that can be used to forecast stock market. From model verification using a nine-year period of Taiwan stock market index (TAIEX) as experimental datasets, it is shown that the proposed model outperforms three listing fuzzy time series (Su et al.,2010; Huarng and Yu ,2006; Chen,1996), and statistic time series models (AR(1),AR(2) and ARMA(1,1)).

Key words: Stock forecasting, price-pattern, time series, fuzzy time series.