Uncovering networks amongst stocks returns by studying nonlinear interactions in high frequency data of the Indian Stock Market using mutual information

TL;DR

This study uses mutual information to detect nonlinear interactions among stocks in the Indian Stock Market during 2014, revealing sector-based clusters and differences in network structure across election periods.

Contribution

It introduces a nonlinear mutual information approach to uncover stock networks and compares its effectiveness with traditional correlation methods in a high-frequency Indian market dataset.

Findings

Mutual information outperforms correlation in network detection.

Identified stable financial sector cluster unaffected by elections.

Detected energy sector as a distinct cluster during all periods.

Abstract

In this paper, we explore the detection of clusters of stocks that are in synergy in the Indian Stock Market and understand their behaviour in different circumstances. We have based our study on high frequency data for the year 2014. This was a year when general elections were held in India, keeping this in mind our data set was divided into 3 subsets, pre-election period: Jan-Feb 2014; election period: Mar-May 2014 and :post-election period: Jun-Dec 2014. On analysing the spectrum of the correlation matrix, quite a few deviations were observed from RMT indicating a correlation across all the stocks. We then used mutual information to capture the non-linearity of the data and compared our results with widely used correlation technique using minimum spanning tree method. With a larger value of power law exponent {\alpha}, corresponding to distribution of degrees in a network, the nonlinear method of mutual information succeeds in establishing effective network in comparison to the correlation method. Of the two prominent clusters detected by our analysis, one corresponds to the financial sector and another to the energy sector. The financial sector emerged as an isolated, standalone cluster, which remain unaffected even during the election periods.