Predicting financial distress in Indonesian life insurance companies with classification methods and synthetic features generation

Financial problems in life insurance companies can become serious if not addressed immediately. Companies experiencing financial distress, for instance, are unable to meet their obligations to pay their liabilities. A company can be categorized as experiencing financial distress when it has an RBC ratio of less than 120%—based on regulation by the Indonesian Finance Service Authority—or ROA < 0 (suffering loss). Therefore, financial distress prediction is carried out to assess the company's current financial condition so that it can be handled early. In this study, we aimed to predict financial distress of Indonesian life insurance companies. We utilized the Support Vector Machine (SVM) classification method, Generalized Extreme Value Regression (GEVR), and Extreme Gradient Boosting (XGB) and by incorporating synthetic feature generation in variable selection. The results of financial distress prediction obtained the best model that can predict the financial condition of life insurance companies in Indonesia at each size, where for sizes 0 and 1, the XGB model with variable selection produces accuracy values of 98.00% and 94.10%, respectively, and AUC values of 100% and 87.40%. Then, at size 2, we can use Stepwise Generalized Extreme Value Regression with accuracy and AUC results of 90.20% and 82.60%, respectively. Each addition of size to the time window classification results tends to reduce the model's performance in predicting the financial condition of life insurance companies in Indonesia.