Abstract

     Financial data, such as financial statements, contain valuable and critical information that may assist stakeholders and investors in optimizing their capital to maximize overall economic growth. Since there are many variables in financial statements, it is crucial to determine the causal relationships, that is, the directional influence between them in a structural way, as well as to understand the associated accounting mechanisms. However, the analysis of variable-to-variable relationships in financial information using standard correlation functions is not sufficient to unveil directionality. Here, we use the volatility constrained correlation (VC correlation) method to predict the directional relationship between two arbitrary variables. We apply the VC correlation method to five significant financial information variables (revenue, net income, operating income, own capital, and market capitalization) of 2321 firms listed on the Tokyo Stock Exchange over 28 years from 1990 to 2018. This study identifies which accounting variables are influential and which are susceptible. Our findings show that operating income is the most influential variable while market capitalization and revenue are the most susceptible variables. Surprisingly, the results differ from the existing intuitive understanding suggested by widely used investment strategy indicators, the price–earnings ratio and the price-to-book ratio, which report that net income and own capital are the most influential variables affecting market capitalization. This analysis may assist managers, stakeholders, and investors to improve financial management performance and optimize firms’ financial strategies in future operations.

Introduction

     The advent of information technology has made it possible to store and classify a large amount of financial data in real time on an unprecedented scale [1]. Financial information can be represented by a vast variety of data types. However, one of the most abundant types of data is multivariate time-series data, in which correlations among different assets are often used to perform risk management analysis, determine investment strategies, and improve financial management. The correlation structure in financial markets has been investigated in many studies at different scales. Preis et al

     analyzed 72 years of daily closing prices of the 30 stocks of the Dow Jones Industrial Average and found that the average correlation among the stocks is linearly scaled with market stress at various time scales [2]. Laloux et al. applied random matrix theory to understand the statistical structure of the correlation matrix of price changes and showed that there is a significant agreement between theoretical predictions and real data related to the density of eigenvalues [3]. This study raises serious questions about Markowitz’s naive use of empirical correlation matrices for portfolio risk management. Plerou et al. also applied random matrix theory to analyze the cross-correlation matrix of price changes of the largest 1000 US stocks [4]. Their findings showed that randomness cannot explain the universal and non-universal properties that enable the identification of cross-correlations between stocks. Jiang and Zheng uncovered the positive and negative subsector structures of Chinese, US, and global stock markets by applying a random matrix theory analysis and taking into account the sign of the components in the eigenvectors of the cross-correlation matrix [5]. Cross-correlations between volume change and price change of financial indexes were studied by [6]. Long-range magnitude cross-correlations in real-world data from various fields, such as finance, have also been studied intensively, using time-lag random matrix theory [7]. The integration of random matrix theory and network methods was also used to unveil correlation and network properties of 20 financial indexes [8]. The importance and novelty of physical methods applied to the study of financial cross-correlation analysis have been highlighted by Kwapien and Drozdz in their extensive review [9]. In particular, the application of random matrix theory to financial multivariate time series data deserved special attention. In this regard, several works have extended and further applied this statistical mechanics approach to the field of finance, namely [10–۱۵].

Conclusion

     Previous studies of correlation coefficients in finance data have found out about the correlation between two variables but not their direction [2,4–۸]. However, we were able to capture the directionality between variables in financial statement data that cannot be captured by ordinary correlation coefficients.

     In this study, the VC correlation approach unveiled the directionality between five major accounting variables, which is difficult to obtain using standard correlation methods. Our data-driven computations yielded new insights on major accounting variables, which can be translated into novel recommendations for investment strategies. We summarize our findings as follows.

     First, from the directionality network, we observed that operating income is the origin of influence on the other four accounting variables (net income, own capital, market capitalization, and revenue). Market capitalization and revenue are the most susceptible accounting variables. Second and more importantly, although market participants often focus on net income and own capitalization to evaluate the share price for investment strategy, operating income may be a better accounting variable on which to focus. Third, the influence order of revenue, operating income, and net income differs from the order of accounting calculation of income statements.