Research On Application Of Resemble Learning Method In Listing Corporation Earnings Forecast

At present, forecast has been an important discipline. With the development of forecast, the application of nonlinear forecast method has been a hot topic in the current research. Researchers have paid close attention to ensemble learning method increasingly, which is one of the four subjects in the machine learning. As an effective machine learning method, ensemble learning has been widely used in various fields, such as corporation finance distress, exchange forecast and text mining.Investors always pay strong attention to the profitability of a corporation. Earnings forecast has been the topic of the current research. Accurate or not of the earnings forecast is associated closely with the interests of investors. Therefore, in order to make more accurate forecast, many kinds of predicting methods had been used in earnings forecast. However, studies show that results based on the recent methods applied in earnings forecast are not very accurate. So, how to forecast earnings in a new and effective method is one problem that needs to be solved urgently.Base on the analysis of the literatures related to earnings forecast, this paper points out the main problems of the existing studies. Then, this paper analyzes the mechanism of earnings forecast by using ensemble learning. When noticed that the earnings forecast is an complicated nonlinear problem, this paper applies ensemble learning models by taking decision tree and BP neural network as basis classifier respectively to forecast the future states of earnings.The last chapter does an empirical study by using the built ensemble learning model. The data used in this paper was the annual report data of the listing companies on Shanghai Stock Exchange and Shenzhen Stock Exchange during the fiscal year of 2001 and 2007. This paper takes the data from the annual report of year 2007 as the test data. After removing those companies which listed between year 2002 and 2007 or companies whose financial variables have many missing values, this paper do a preprocessing to the resulting data. The preprocessing contains imputing miss values, dealing with outliers. Moreover, in order to fit the characteristic of BP neural network, data standardization is done. In order to be more practical and to give more guides to investors, the dependent variable, namely, earnings per share, has been classified into three classes. After built the basis classifiers, by training the models on the dataset contains the selected financial variables and target variable, this paper predicts the future states of earnings in year 2007 by using the decision tree ensemble learning and neural network ensemble learning respectively. The results show that the forecast accuracies of the two ensemble learning methods are more robust than the individual decision tree model and the neural network model.