Stock Price Forecast Based On Candlestick Chart

Technical analysis uses historical price information to form charts and release trading signals to help investors make investment decisions.As an analysis tool for many investors at home and abroad,it has formed many theories and methods after years of summary and practice.However,technical analysis relies too much on experience summarization,and its results contain certain concerns.Therefore,controversy over the effectiveness of technical analysis methods has always existed.As a result,whether the effective information is really contained in the K-line chart has become the core of people’s attention,and has also become the research object of some scholars.In recent years,deep learning research has been widely carried out.Among them,the convolutional neural network has developed more maturely in the field of visual recognition.Therefore,in the research of this paper,the unique advantages of convolutional neural networks in image recognition are combined with K-line graph fine-grained image recognition.Based on B-CNN(Bilinear-CNN,dual-stream linear convolution network),the binary tree linear Convolutional neural network model and the multi-stage supervised dual-stream linear convolutional network model are proposed.And they are tested the accuracy of candle statistical chart,MACD statistical chart,volume bar chart statistical chart,KDJ / AR(BR)broken line statistical chart.The results are 58.41%,56.85%,56.56%,54.4% respectively.Finally,through the fusion algorithm,the comprehensive accuracy rate reached 58.85%.In the end,the research of this thesis found that it is feasible to study the K-line graph directly as the input form of the convolutional neural network.And preliminary obtained,the K-line diagram does contain the conclusion that can predict the future price changes,that is,from the perspective of the research of this paper,the graphic technical analysis is effective.