Automated Generation and Verification of Stock-Price Technical Analysis Report

Abstract

Technical analysis of stock prices is conducted depending on the experience of the market participants, and the trends in the prices are often predicted using specific patterns of the stock price movements represented in the stock price chart. However, it is not easy for inexperienced traders to perform such analyses, and it is also time-consuming and costly for professional chart analysts to analyze every chart. In this study, we developed a deep neural network model to automatically generate technical stock analysis reports for a given stock price chart. To this end, we designed an encoder to capture both the long-term and short-term characteristics of the stock time-series data. The encoder consists of several modules, each of which receives stock data of different lengths from the same reference date. The long stock time-series is summarized using a 1D convolutional neural network, and the features are extracted using an LSTM. Based on the extracted features, a stock technical analysis report is generated. We evaluated the quality of the generated sentences and performed a case study to verify if the generated analysis report was reliable. Through visualization experiments, we confirmed that the proposed model can correctly capture the characteristic patterns represented in a stock chart.