Critical elements in the process of predicting corporate bankruptcies are selection of financial indicators and construction of a forecast model. Many traditional methods implement these two tasks separately, and do not guarantee the optimality on the whole process. This study uses the RealAdaBoost algorithm so that the selection of indicators and the construction of a model can be realized within a single coherent framework in the process of forecasting bankruptcies. RealAdaBoost iteratively chooses a financial indicator and the weight of learning samples that are mistakenly classified by the indicator increases in the next iteration step, and because of this, a combination of indicators that complement each other is naturally extracted. Setting financial ratios generated from the balance sheets and profit-and-loss statements of 150 failed firms and 150 continuing firms as candidates of indicators, our proposed method selects the following four ratios in most cases: 1) Net income before taxes and other adjustments / Current liabilities, 2) Total investments and other assets / Capital stock, 3) Total liabilities / Fixed assets, and 4) Owned capital / Capital stock. Discrimination of RealAdaBoost with four financial ratios shows an identification rate of 0.893 under the leave-one-out cross validation.
|Number of pages||14|
|Publication status||Published - Jun 2016|
- Bankruptcy prediction
- Financial ratios