2020 Fiscal Year Final Research Report

Financial Market Analysis with Deep Learning -Stock market emotion extraction and monetary policy-

Research Project

PDF

Project/Area Number	17K01267
Research Category	Grant-in-Aid for Scientific Research (C)
Allocation Type	Multi-year Fund
Section	一般
Research Field	Social systems engineering/Safety system
Research Institution	The University of Kitakyushu
Principal Investigator	Ikeda Yoshikazu 北九州市立大学, 経済学部, 教授 (10334880)
Co-Investigator(Kenkyū-buntansha)	林田実北九州市立大学, 経済学部, 教授 (20198873)
Project Period (FY)	2017-04-01 – 2021-03-31
Keywords	株価変動予測 / 畳み込みニューラルネットワーク / ディープラーニング / 株価ローソク足チャート / 出来高グラフ / アンサンブル学習
Outline of Final Research Achievements	We have shown that the convolutional neural network, which has been applied to category prediction by extracting features from images, can predict stock price fluctuations with a high enough accuracy compared to the random prediction accuracy of 50%. In addition, it was shown that the accuracy of the prediction was improved compared to the prediction by a single model by adopting an ensemble, which is a majority vote of multiple models, instead of a single convolutional neural network model. In order for the majority voting to work effectively, we proposed a method of adding diversity to each predictor. These majority decisions can be seen as one of the emotions of the market.
Free Research Field	複雑系経済学
Academic Significance and Societal Importance of the Research Achievements	金融市場などを発端とする金融インパクトは金融市場の不安定さをもたらし，実体経済へも多大なる影響を与えることとなるので、これら変動を予測し，抑制することは安定した経済発展にとって重要な課題であると考えられる．本研究では，機械学習の手法を用い，金融市場変動における価格変動予測や多数の市場の多数の参加者の多様性（これは雰囲気（強気，弱気）など感情に類似のものととらえることもできる）を再現し、一種の市場参加者の感情推定を可能にすることを示すことができたと考える。