The Research On IPO With BP Neural Network

Low risk, high returns, great oversubscription and high turnover of Initial Public Offering are always obvious features of Chinese capital market, which damage market efficiency and resource allocation efficiency. So, the problem that the difference between price of IPO and price in expected secondary market becomes one important part of our issue market system. Since 1994, our government had reformed our security primary system, firstly realized the conversion from examination and approval system to checking system in April 2001, and then at first time adopted initial public offering of trial request system on Jan 1, 2005. But the reformation of offering systems can not fundamentally improve our primary market efficiency, so on this issue still needs to study thoroughly.At present, Discount cash flow model, Economic value added model and Comparable company analysis model are mainly IPO pricing models, and Marketing returns model, Real options valuation model and Multiple factors pricing model support to main models. But in practice, because of insufficient information and inaccurate profit forecast, these models are not exact and operational. So we need to design the model avoiding these adverse effects in those IPO pricing models, in order to reduce the estimation errors.By contrast, artificial neural network is a kind of intelligent data-processing methods, which is good at dealing with non-linear data relationship, so it shows its superiority in complex system. And as a kind of artificial neural network, BP neural network does good job at forecast and evaluation in condition of insufficient information with high-speed computing, learning ability and approaching any non-linear continuous function in theory. So the IPO pricing model based on BP neural network is able to avoid the shortage of information and subjective judgment. Then, the principle of the IPO pricing model based on BP neural network firstly analyses the influencing factors of IPO pricing, then collects relative information of listed companies and find the relationship between input variables and output through simulating the process in BP neural network, in order to get the pricing model's program, finally obtain the evaluation of other newly listed companies' price by putting other newly listed companies' information to the program. Above procedure are realized in Matlab7.1, and take example of comparing these IPO pricing models. As a result, the IPO pricing model based on BP neural network can make up for inadequacy of other models and to some extent improve valuation accuracy, and give the help of enhancing prime market efficiency.