The Modeling of Earnings Prediction by Time-Delay Neural Network
Although the use of earnings prediction in supporting investment decisions has been prevailing in practice, an accounting-based analysis for modeling the key accounting components by time-delay machine learning technique is unexplored. Traditional time-series techniques fail to handle complex data structure, and the fundamental analysis approach cannot model multiple periods’ data effectively. Thus, this study aims to explore the crucial relationships among future earnings and the main historical accounting components, i.e. cash-flow and accrual components. The research method leverages the flexible learning capability of artificial neural network (ANN) with time-delay data structure. The major findings suggested that adding accrual components is helpful for better earnings prediction, and the proposed 5-period time-delay ANN model may capture the future earnings in a positive way. The results of this study may help to support investment decisions and better understanding for the role of accruals in earnings.