Adaboost And Bagging Ensemble Approaches With Neural Network As Base Learner For Financial Distress Prediction Of Chinese Construction And Real Estate Companies

In recent years, as China’s sustained, healthy and rapid development of the national economy, the construction and real estate industries have developed rapidly, its great potential, high correlation, driving force and other characteristics appeared in China’s economic development process, and gradually become basic industries related to the people’s livelihood and they play an important role in the macro-economic development. Real estate as an emerging industry in China, it’s a capital-intensive industry that needs a long project period, a large amount of investment, and a long pay-back period among others. In addition, the construction and real estate companies have more social risks in the process of operational globalization. After the global financial crisis of2008, many countries have regulated the construction and real estate industry to a large degree. In this regard, construction and real estate companies, particularly those small ones with precarious financial station, are confronted with even greater financial risk of bankruptcy or being acquired. Therefore, the establishment of an effective method of real estate industry’s financial crisis early warning system is of great theoretical and practical significance.This article to listed real estate companies in Shanghai and Shenzhen stock market as a research object, collection time interval in2000-2010for two consecutive fiscal years of losses or net assets per share lower than the face value of the shares was the special treatment (ST) sample of the real estate business as a financial crisis. With unconventional pairing of extraction methods, eventually selecting the real estate industry in Shanghai and Shenzhen stock market. There were85sample companies, of which ST companies33and other52are non-ST companies. Building a financial indicator system with35indicators that involve solvency, profitability, operational capacity, per-share ratio, operating cash flow ratio, structural ratio, and growth capacity. In addition, t-test, correlation analysis and stepwise discriminant analysis method are utilized respectively to obtain the final variables for FDP modeling. This article uses back-propagation neural network (BP_NN) as the base learner and constructs two classifier ensemble models, BPNN_AdaBoost and BPNN_Bagging, based on AdaBoost and Bagging ensemble learning methods, for empirical study on China’s construction and real estate companies. Results of BPNN_AdaBoost and BPNN_Bagging models are comprehensively compared with those of the single BP_NN model and classical Z3Score model.Empirical studies have shown that in the prediction of a three-year, the accuracy of BPNN_AdaBoost and BPNN_Bagging model is higher than a single BP neural network and classical Z3_Score model. On the significance level of5%, whether BPNN_AdaBoost model or BPNN_Bagging model are significantly superior to a single BP neural network in the prediction in T-1, T-2and T-3years, improving remarkably. While experimental results we can see that BPNN_Bagging model is more suitable for short-term and medium-term FDP in one-or two-year advance, and BPNN_AdaBoost model in the long term have a stronger generalization, classification forecasting can be more effective.