Growth Enterprise Value Assessment Based on RF-BP Neural Network

Abstract

Enterprise value assessment is an important prerequisite for company activities including listings, mergers, and acquisitions. In order to overcome the limitations of the methods including income method, market law, and to improve the accuracy of enterprise value assessment, this research has used the historical data of 164 growth/developing companies listed on the Growth Enterprise Market (GEM) to conduct neural network training, so as to explore the accuracy and applicability of the evaluation methods of enterprise value based on deep learning. The most important five types of indicators from many indicators were firstly selected with the random forest (RF) algorithm, then used as neural network input neurons. Meanwhile, in order to improve the learning ability of the neural network, this study has used RMSProp algorithm to rectify the node weights and offsets, which reduces the problem that the neural network may fall into the local minimum. The results gained from the training and testing of the data show that the method has high accuracy.