Statistical analysis of financial data with many zeros

One of the core issues in finance is to understand why firms finance themselves as they do. This issue has become increasingly important because how firms are financed influences their performance and value. Since the 1950s, the capital structure literature has addressed this fundamental issue by focusing on a firm's mix of debt and equity. However, firms often use more than one type of debt claim. Furthermore, some firms use certain types of debt claim that others do not use. The financing choices of various forms of debt claim and different amounts of debt issued lead to financial data with multiple continuous proportions and many zeros implied by debt structures. We propose a novel method for addressing such choice‐implied statistical issues. Our method is based on choice probability‐driven submodels and is empirically implementable even in large dimensions. Its performance is demonstrated by simulations. Its application to the analysis of debt structures of U.S. corporations reveals that the determinants of the choice to use a particular form of debt and how much of that type of debt to use are not identical: a valuable insight missing from prior financial research. Our methodology is applicable not only to firm financing decisions in corporate finance, but also to choices in other critical areas of finance such as household investment decisions in household finance. WIREs Comput Stat 2017, 9:e1399. doi: 10.1002/wics.1399 This article is categorized under: Applications of Computational Statistics > Computational Finance Statistical and Graphical Methods of Data Analysis > Multivariate Analysis