High-frequency Financial Time Series Analysis Based on BP Neural Networks

Paper

http://www.lin-baobao.com/static/files/graduate_paper.pdf

Video for Demonstration

http://www.lin-baobao.com/static/videos/graduate_project.flv

Application

http://www.lin-baobao.com/bp_finance/php/login/

This project includes three sections.

• GetData

  Data crawling.

  Responsible for crawling and processing the high frequency data of stock transcation in recent years. And the data format would be converted to an appropriate format.

• predict

  Core algorithm.

  Implement the Back Propagation Neural Networks without using any framework (all code is completely written by myself).

• php

  System integration.

  Integrate the first two parts into a system.

Project description

It included data crawling, model design, code implementation, result analysis, experiments and system integration.

• 1. Crawled the data of some stocks’ transactions per day.
• 2. Chose BP Neural Network as model, deduced the formulas myself and implemented code without using any framework.
• 3. The direction accuracy was about 55%-60% and the relative error between the actual and the predicted price maintained at 0.015%. Reached the conclusion that high-frequency data is better for stock price prediction via testing data’s impact at different frequencies.
• 4. Integrated the algorithm into a real-time stock price prediction system.

Deployment

Need to deploy mysql, and the sql file is in finance.sql

Ways to run

Just call the run.py directly.