Applied Research Of K-means And Non-homogeneous Hidden Markov Models In The Fluctuation Trend Of Stock-price

There are many methods to predict stock tendency.The traditional time series method only considers the relationship between stock data and time and fails to pay attention to the characteristics of stock price sequence itself,such as variability,linkage and complexity.However,the machine learning algorithm only focus on the trait of the stock price sequence itself of machine learning,thereby forecast the stock price.This paper adopted the advantages of the two above methods,combined with Hidden Markov Model(HMM),changed the homogeneity,added to the impact of macroeconomy,and made up the non-homogeneous hidden Markov model(NHMM)to estimate stock price trend,according to the characteristics of Chinese stock market.In this paper,the macroeconomic factors are added to the forecasting model,and the impact weight of the macroeconomics on the stock price trend is considered by means of PCA-BP,then the non-homogeneous hidden Markov model(NHMM)is constructed to predict the stock price trend.On the one hand,it eliminates the influence of information redundancy between macroeconomic factors,on the other hand,it combines macroeconomics and proposes a non-homogeneous hidden Markov model(NHMM).It selected 2960 items of data with closing stock price from July in 2006 to August in 2018 and made up the model to predict using the trend of closing price for every 20 days as a sample.At the same time,we thought about nine macroeconomic factors including MEI(Entrepreneur Confidence Index),CPI,IVA(Industrial Increase)etc.and also considered the addition way of PCA-BP macroeconomic factor which affect weight in different states so that constructed non-homogeneous hidden Markov Model(NHMM).This paper is divided into three steps to build and demonstrate the model.First of all,K-means clustering is used to cluster all the 20-day closing price trends,and four trend types are clustered as the observation sequence of the non-homogeneous hidden Markov model(NHMM).Secondly,the weight was affected by the addition way of PCA-BP macroeconomic factor,specifically,first reducing dimension of the selected macroeconomic factors by PCA and then the results as the input layer of BP neural network,finally making up NHMM model by the macro economy which acted as a weight change to NHMM transition probability.The last step is to compare the prediction results of HMM and NHMM model.The empirical results showed that there was indeed a large non-linear relationship between macroeconomic and stock price changes.The macroeconomics influence the stock market's direction.The prediction accuracy of the non-homogeneous hidden Markov model(NHMM)is higher than counterpart of the hidden Markov model(HMM),with s74.07%and 51.85%respectively.