Change in hierarchy of the financial networks: A study on firms of an emerging market in Bangladesh.

We investigate the hierarchical structure of Dhaka stocks' financial networks, known as an emerging market, from 2008 to 2020. To do so, we determine correlations from the returns of the firms over a one-year time window. Then, we construct a minimum spanning tree (MST) from correlations and calculate the hierarchy of the tree using the hierarchical path. We find that during the unprecedented crisis in 2010-11, the hierarchy of this emerging market did not sharply increase like in developed markets, implying the absence of a compact cluster in the center of the tree. Noticeably, the hierarchy fell before the big crashes in the Bangladeshi local market, and the lowest value was found in 2010, just before the 2011 Bangladesh market scam. We also observe a lower hierarchical MST during COVID-19, which implies that the network is fragile and vulnerable to financial crises not seen in developed markets. Moreover, the volatility in the topological indicators of the MST indicates that the network is adequately responding to crises and that the firms that play an important role in the market during our analysis periods are financial, particularly the insurance companies. We notice that the largest degrees are minimal compared to the total number of nodes in the tree, implying that the network nodes are somewhat locally compact rather than globally centrally coupled. For this random structure of the emerging market, the network properties do not properly reflect the hierarchy, especially during crises. Identifying hierarchies, topological indicators, and significant firms will be useful for understanding the movement of an emerging market like Dhaka Stock exchange (DSE), which will be useful for policymakers to develop the market.

Effect of network size on comparing different stock networks.

We analyzed complex networks generated by the threshold method in the Korean and Indian stock markets during the non-crisis period of 2004 and the crisis period of 2008, while varying the size of the system. To create the stock network, we randomly selected N stock indices from the market and constructed the network based on cross-correlation among the time series of stock prices. We computed the average shortest path length L and average clustering coefficient C for several ensembles of generated stock networks and found that both metrics are influenced by network size. Since L and C are affected by network size N, a direct comparison of graph measures between stock networks with different numbers of nodes could lead to erroneous conclusions. However, we observed that the dependency of network measures on N is significantly reduced when comparing larger networks with normalized shortest path lengths. Additionally, we discovered that the effect of network size on network measures during the crisis period is almost negligible compared to the non-crisis periods.

Application of principal component analysis on temporal evolution of COVID-19.

The COVID-19 is one of the worst pandemics in modern history. We applied principal component analysis (PCA) to the daily time series of the COVID-19 death cases and confirmed cases for the top 25 countries from April of 2020 to February of 2021. We calculated the eigenvalues and eigenvectors of the cross-correlation matrix of the changes in daily accumulated data over monthly time windows. The largest eigenvalue describes the overall evolution dynamics of the COVID-19 and indicates that evolution was faster in April of 2020 than in any other period. By using the first two PC coefficients, we can identify the group dynamics of the COVID-19 evolution. We observed groups under critical states in the loading plot and found that American and European countries are represented by strong clusters in the loading plot. The first PC plays an important role and the correlations (C1) between the normalized logarithmic changes in deaths or confirmed cases and the first PCs may be used as indicators of different phases of the COVID-19. By varying C1 over time, we identified different phases of the COVID-19 in the analyzed countries over the target time period.