Research On Forecast Model Based On Cycle And Trend Decomposition

The futures market is an indispensable part of the financial market,and research on futures prices is of great significance for commodity price discovery and market stability.This thesis takes glass futures price as an example,selects the weekly closing price data of the main contract from December 3,2012 to August 26,2022,and uses the cycle and trend related research in macroeconomics and signal decomposition theory to transfer it to the futures market to construct a prediction model.As a bulk commodity,before making predictions,seasonal adjustment is applied to glass futures price to eliminate the interference of seasonal effects.After adjustment,the glass futures price is decomposed into cycles and trends.Six decomposition methods are used in this study,including model-based decomposition methods-BN decomposition and UC decomposition,filter-based decomposition methods-HP filtering and CF filtering,and multi-period decomposition methods-wavelet decomposition and empirical mode decomposition.Then,LSTM is used to predict the cycles and trends separately,and the prediction results are added together to obtain the result of the decomposition prediction model.Different decomposition methods extract different information,resulting in differences in the decomposition prediction model.Then,based on the characteristics of a single decomposition prediction model and drawing on the "momentum effect" in stock returns,the concept of "continuity of effectiveness evaluation sequence" was proposed.It is believed that if a prediction method is accurate in the current period,it is likely to be accurate in the next period as well,and relevant proof steps are given.Based on the discovered "effect evaluation sequence continuity",two variable-weight combination prediction models were constructed.It was found that the constructed combination prediction models had good predictive ability and improved the accuracy of predictions.Finally,the glass futures price case was extended to the general situation and simulated experiments were conducted.The experiment found that the decomposition prediction model had better predictive performance than the non-decomposition prediction model.In addition,the attribution of high-frequency components in the decomposition would affect the predictive performance of the decomposition prediction model.At the same time,this thesis proposes a decomposition criterion for the decomposition prediction model,that is,under the premise of not losing all information,the decomposition should make the cycle or trend as smooth as possible to obtain a better decomposition prediction model.In summary,this thesis combines academic research with empirical research and proposes a prediction model that can serve as a powerful tool for predicting glass futures price while providing new ideas for the decomposition prediction research of financial markets.