A topological based feature extraction method for the stock market

Abstract

We proposed a topology-based method for pre-processed time series data extracted from stock market data. The topology features are extracted from data after denoising and normalization by using a version of weighted Vietoris-Rips complex. We compare the features from bullish, bearish and normal periods of the Chinese stock market and found significant differences between the features extracted from the groups. Based on the previous research mentioned in the context, we proposed a topology-based stock market index which has the ability to distinguish different stages of the stock market and forewarn stock market crashes.

Keywords

• topological data analysis
• stock market
• feature extraction