This study examines the role of comprehensive income and its components, in addition to net income, as inputs to forecast bankruptcy. Using a matched sample of 466 (233 pairs) U.S. bankrupt and non-bankrupt firms from 1993 to 2014, we build a bankruptcy prediction model using random forest classification. Compared with the benchmark model, our proposed model’s accuracy increases by 1.5% and the Type I error decreases by up to 3%. A variable importance analysis reveals that comprehensive income is consistently the most useful variable for bankruptcy prediction. A variable interaction analysis shows that the top interaction pair includes one Altman variable and comprehensive income. Finally, we analyze bankrupt firms that our model identifies but the benchmark model misclassifies; we find that such firm’ other comprehensive income is consistently negative, suggesting that firms’ macroeconomic risk exposure plays a key role in bankruptcy prediction.
|出版狀態||已被接受 - 2022|