Many machine learning techniques such as deep learning models need tons of data to train but in reality the financial data are limited and then restrain the power of models. Further, many investment funds need more new data to test their trading strategies and overcome overfitting problems. In the past, many time series models like ARIMA and GARCH can generate new data , but these mehtods rely on strong model assumption. In comparison, we try to implement the state-of-art GANs method, which is a game-thoery based unsupervised learning model, to generate new financial data.
In this project, we tried to implement several GANs techniques to simulate open, high, low, and close price of three share price index futures in China and generate one-minute charts for day trading (240 minutes a day since 2016-01-01). Our main network structures are DCGANs, and we restructure the raw dataset using a log-return trick, which is essential for GANs to learning continuous time series features.
The three datasets below are provided by likelihood lab, which is a non-profit AI research lab.
此处可能存在不合适展示的内容,页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容,可点击提交进行申诉,我们将尽快为您处理。