Article Preview
Top1. Introduction
Stocks are the investment of money in a corporation, that represents the ownership of the stockholder within the corporation (Chmielewski et al., 2020). Nowadays stock markets have a lot of fluctuating data that needs to be analyzed by investors and managers (Rahimnezhad et al., 2020). It is assumed that in the stock market future value of a company's share can be predicted so that investors can buy low and sell high to make a profit (Patil et al., 2020). If a person has the knowledge to make financially responsible choices, he can predict the outcome of stocks and decide whether to sell or buy the stocks, and can enjoy the benefits of growing wealth (Chmielewski et al., 2020). A stock study can be started by analyzing the correlation between the stocks and the correlation can be found from the fluctuations recorded in the prices of shares (Purqon & Jamaludin, 2021). Correlation-based stock networks can be used to determine the structure of complex share markets (Zhao et al., 2021). Correlation is bivariate in nature, so a high correlation between the stocks in indices represents that the value of these stock’s price moves together (going up or down), over time (Rahimnezhad et al., 2020). This dependency will help the trader to find out the stocks belonging to a community, which is moving in upward direction. Clustering the stock correlation network into communities is an important technique for market analysis as it reveals important information related to trends and risk diversification in stock markets. (Chmielewski et al., 2020; Wu et al., 2015; Zhao et al., 2021).
Detecting communities may provide a more explainable summary of many complex systems (Hoffmann et al., 2020) and complex systems that exist in real-life or in society can be represented in terms of networks (Colliri & Zhao, 2021) such as transportation networks, computer networks, social network, biological network, etc. Complex network theory also provides simple and useful methods to explore the characteristics of many complex systems such as molecular activities of cells in the human body, and communication between people over social media (Gui et al., 2014). Over time the complex networks have attracted many researchers in different fields including economics and finance (Wu et al., 2015). Complex network theory also provides us with a new approach to studying stock markets. In this paper, a stock correlation network is created based on the closing price of stocks, as the price is the core reflection of stocks in the market (Gui et al., 2014) and complex network theory is used to study the characteristics of community structure in the network.
The motivation to perform this work comes from the fact that most of the time the fluctuation in the price of stocks is not independent of each other and there exist complex dependencies between them. So, it can be said that the performance of the companies in markets is correlated and the calculation of this correlation can be a way to show the dependencies between the stocks (Rahimnezhad et al., 2020). There can be two factors because of which, this correlation exists, first the general direction of the market, second cyclicity in similar segments of the stock markets (Alshahrani et al., 2015). In this work, the stock correlation network of four stock indexes (Nifty200, FTSE100, Nasdaq100, ASX200) is created using the closing price of stocks. A stock correlation network is a graph where nodes represent the stocks or companies in the index and the edge represents the relation between the stocks. The relationship is constructed by evaluating the correlation coefficient between the pair of stocks. After the creation of the network graph BAT optimization algorithm is used to find out the community structure using modularity as the fitness function. The reason to use the BAT algorithm for this problem comes from (Song et al., 2016), where the authors have shown that the BAT algorithm performs better than particle swarm optimization (PSO) and spectral clustering algorithm for community detection. In this paper, the following research contributions are made: