Adapting Altman's bankruptcy prediction model to the compositional data methodology

Battle Count: 67

The paper is primarily focused on corporate bankruptcy prediction and credit risk assessment rather than quantitative trading. However, the compositional data methodology and log-ratio transformations could be relevant for factor investing, credit risk modeling in portfolio management, and financial distress signals used in equity strategies. The finding that CoDa methods improve sensitivity (recall) for bankrupt firms is relevant for risk management in trading portfolios. The methodology is more applicable to fundamental analysis and credit scoring than to high-frequency or algorithmic trading.

The methodology is moderately complex. It requires understanding of compositional data theory, log-ratio transformations, and proper handling of zeros in accounting data. However, the actual implementation uses standard R packages (glm, class, randomForest) and CoDaPack. The main complexity lies in correctly constructing the pairwise log-ratios, handling zero imputation, and understanding when to use D-1 vs D(D-1)/2 plr depending on the method. The paper provides clear formulas and a connected acyclic graph for interpretation.

The paper uses R software with standard packages (glm, class, randomForest), CoDaPack for zero imputation, and publicly accessible SABI database. The methodology is well-documented with specific parameters (100 trees, k=5, 70/30 split, downsampling). However, the curated data is only available upon reasonable request from the corresponding author, and the exact random seeds for train/test splits and downsampling are not specified.