Research On Peer Group Stock Market Trend Predicting Algorithm Based On Deep Computing

Peer group algorithm is an unsupervised learning technique,it can overcome the defect of traditional supervision learning technique,which are unable to effectively detecting new stock market patterns.But the quality of the peer group algorithm is poor and could not predict the movements of the stock market.This dissertation has indicated the disadvantage of PG algorithm,and discussed the significance of presenting a new peer group generation method,which based on depth computing.First,calculating the similarity of band between target stock and candidate stock,and then depth calculation based on intimacy,dependency and liveness.At last,generating peer group of the target stock,and proving the quality of peer group generated by depth calculation was superior to the PG algorithm.Because of the unpredictive of the PG algorithm,a new algorithm(DPG-AR)was put forward,which combining peer group algorithm and AR model.DPG-AR algorithm generates peer group by using depth computing,updating the weights of peer group members,calculating the time series of the peer group closing quotation and then predicting the target stock trend by using auto regression model.Then,the model optimization method was discussed.At last,by contrast experiment on Shanghai composite index and the corresponding stock,and the simulation demonstrated the better performance of the DPG-AR algorithm.In the DPG-AR algorithm,there were also have limitations,the peer group's select time was fixed,and the relationship between volume and price leads to future differences in the trend has not been considered.Therefore,associate stock trend with volume,presenting a peer group whose time range was unlimited generate algorithm(DVTPG)which was integrated into the volume-price relation.First,Choosing stocks which in the same trend of target stock price graph according to band range and intimacy.Then,according to stock price fluctuate,DVTPG algorithm choose stocks with the small of fluctuate range.And then,in order to eliminate the impact of the noise of the SCI on the correlation between stocks,DVTPG algorithm choose stocks which in the strong correlation by using partial correlation coefficient.At last,the algorithm choose stocks with the same volume-price relation as target stock by using volume fluctuate and generating peer group of the target stock.A new algorithm(DVTPG-AR)was given by combining DVTPG and AR model to predict stock future trend by using historical form and volume-price relation,and the effectiveness was showed in the experiment on Shanghai Composite Index.