Failure prediction for hospitality firms in U.S. and Korea using logit and neural networks models

This study developed failure prediction models for Korean and U.S. hospitality firms using logistic regression and artificial neural networks (ANN) techniques.;For Korean hospitality firms, the one-variable logit model with interest coverage ratio correctly classified 83.74% of in-sample firms and 76.32% of hold-out firms. The ratio's negative coefficient suggests that low interest coverage of a firm increases its failure probability. To prevent the failures, Korean hospitality firms need to move away from heavily leveraged financial structure. The developed ANN model demonstrated an overall classification rate of 86.18% for in-sample firms and 77.63% for hold-out firms. Empirically, this study shows that the logit model is not inferior to the ANN model in terms of prediction accuracy. In addition, the logit model allows its user to interpret the coefficient of each variable and draw practical implications. Therefore, it is recommended to employ the logit model for predicting hospitality firm failures in Korea.;For U.S. hospitality firms, the logit model retained three ratios: earnings before interest, tax, depreciation and amortization (EBITDA) to current liabilities (CL), quick ratio, and debt ratio. These ratios imply that, to decrease the probability of failure, U.S. firms need to: (1) exercise a tight control on the operating costs; (2) increase sales revenue by pursuing market-share gains; (3) invest in operating assets that produce higher returns than cash or marketable securities; (4) adopt a conservative financing policy. The logit model correctly classified 83.33% of the in-samples firms and 77.63% of the holdout firms. The estimated ANN model, on the other hand, demonstrated overall classification rates of 91.98% on in-sample firms and 85% on hold-out firms. While the ANN model may achieve higher classification rates, the downside is the model's lack of self-explanation capabilities. The decision for model selection, therefore, should be made based on the objective of classification. If the primary objective is to classify a set of observations as accurately as possible, then the ANN model may be used. Alternatively, if the researcher wishes to make a practical interpretation of the developed model, then it is recommended to use the logit model for predicting firm failures.