Quantifying Accounting Structure
We develop computational tools, inspired by Claude Shannon's entropy concept, to capture and quantify the graph properties of the classified bookkeeping records.
Image Designed by Freepik
Pattern Recognition and Anomaly Detection in Bookkeeping Data
We introduce the Minimum Description Length (MDL) principle in performing tasks of pattern recognition and anomaly detection in bookkeeping data, leveraging the graph structure of double-entry bookkeeping.
Image Designed by Freepik
Research Themes and Projects
Industry Solutions Research
These research projects are conducted with participants from three broad areas. Industry partners who provides practical research problems, proprietory data, and some funding; 2) Accounting domain experience who frame these problems into research questions related to some fundamental and theoretical accounting structure and guide the data analysis with the accounting theoretical insights; 3) Computer science or other technical research experts from CMU's School of Computer Science or other departments who provide technical solutions to the research questions posted by the accounting domain experts and industry partners. Example of such projects include
Accounting Structure
We theoretically develop and empirically apply a new information measure that quantifies the structural dimension of financial statements. Building on the entropy concept from Shannon's (1948) information theory, our approach captures firm fundamentals directly embedded in accounting numbers and formally measures the information conveyed by their classification structure. The current projects include
- Accounting Classification Entropy (ACE)
- Accounting Graph Entropy (AGE)
Foundational: Three Laws of Double-Entry Bookkeeping
In this position paper, three laws are proposed to describe the conventional double-entry system of bookkeeping: (1) the balance law; (2) the conservation law; (3) the linearity law. Using the three laws, we develop three entropy measures: Balance-sheet Node-entropy, Transaction Edge-entropy, and Bookkeeping Graph (Laplacian) entropy. The graph-theoretic mathematical representation we construct here offers a framework for some recent computational experiments which deploy related entropy and other metrics computed on bookkeeping graphs constructed using financial statement data from publicly traded companies. Our position is that quantifying the bookkeeping graph structure has the potential to offer insights into accounting information transmission, processing, and final end-uses.- March-2025 Version: initial presentation at Columbia Business School (limited distribution)
- April-2026 Version: major addition: proof of uniqueness of the Edge-Entropy measure (PDF)