Stock Price Trend Prediction Research Based On Time Series Similarity

With the expanding of the computer technology application, the concept of time series oftenappears in traffic, business, science, finance and other fields. More and more people are valuing thetime series analysis and processing technology. Traditional time series analysis methods tend tomatch the time series into some mathematical models to analysis and predict its entirety. But manyreal data can not meet the requirements of model parameters. So that Analog Complexing algorithmbased on time series similarity with nonparametric regression characteristic become the focus ofresearch in this field.Based on time series similarity measure method and Analog Complexing algorithm, this paperput forward a kind of unequal length time series similarity measure method, design a time seriestrend prediction scheme with strong applicability and carry out empirical analysis with the realstock price data. The main work is as follows:Firstly, study the time series similarity measure method and study the deformation which oftenappear in the time series such as amplitude translation, scaling, linear drift, timeline scaling etc. Andconclude that good time series similarity measure method should be not sensitive to the deformationabove.Second, further study Analog Complexing algorithm for time series prediction and discuss thenonparametric regression model.Analog Complexing algorithm as a kind of typical nonparametricregression method has good application prospect.Thirdly, put forward Refined Cosine Based Similarity for Unequal Length Time Series, whichis not sensitive to amplitude translation, amplitude scaling, linear drift, timeline scaling etc. On thebasis of the basic cosine formula based, it calculate the similarity of unequal length time seriesthrough equal length processing and normalization processingFourthly, with the use of Analog Complexing algorithm for Time series prediction and RefinedCosine Based Similarity for Unequal Length Time Series which is put forward by this article,design a scheme for time series prediction. This article predicts the stock price trend as experimentin two cases, equal length search window and unequal length search window. The scheme canaccurately predict stock price movement direction，but as to the precise prediction of the futurevalue is not satisfactory, which need further research and improvement.Finally, design a prediction support scheme which provides decision support for users throughthe analysis of stock classic graphics, the maintenance of classic graphics database and the search ofclassical graph in the near history. The scheme is not to predict stock price movements, but bysearching out the classic graphics and provide to users, to support users to predict, and finallyreduce the burden of users.