我国主板证券市场新股发行定价的研究

发布时间：2018-03-02 12:18

本文选题：BP神经网络　切入点：VaR方法　出处：《南京理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着中国证券市场改革的深入,我国主板市场新股发行定价的准确性也愈发重要。由于每一个国家的证券市场都具有不可完全复制性,因此在探索与借鉴的基础上,找到适合我国主板市场的定价方法,寻求合理的股票价格成为本文的核心问题。本文首先研究传统定价方法,并比较它们之间的优缺点。在此基础上,寻找理论与方法的突破,研究发现传统方法已经不适合现有的中国主板市场,而期权、风险、神经网络、博弈等概念的引入给予新股发行定价一个进步的空间。对于主板市场的定价,BP神经网络方法相对适用,它可以使误差率达到15%左右,但是,我们认为,精确的风险评估与因素选取可以使定价的准确性提高,因此我们引入风险评估并对因素进行选取。同时,各国上市环境的差异是造成定价方法不能直接生搬硬套的另一个原因,所以筛选比较具有代表性的美国、新加坡、日本和德国这4个国家,与我国进行上市制度的比较。本文探索BP神经网络与VaR风险定价结合的方法来进行股票定价,并得出CVaR比VaR方法更能准确衡量风险的结论,因为前者误差率较低,平均仅达到12.69%,较以前的BP神经网络方法降低2%以上。同时加入政府制度的因素:将不可量化的政府制度作为上市成本因素,且利用归一化方法统一不同的量纲。在深入研究中国、美国、新加坡、日本、德国这五个国家的上市制度的基础上,利用改进的BP神经网络模型进一步降低误差率至11.96%,使定价更有效,这也是上市时间最短、上市资金最少、监管力度最强的德国政府制度下的定价模型所得到的误差率,同时我们得出监管力度是上市时间、上市资金和监管力度这三个因素中最重要的因素的结论,对中国政府进一步改革上市制度有很大的助益。
[Abstract]:With the deepening of the reform of China stock market, IPO pricing accuracy of the motherboard market shares in China is increasingly important. Because of each country's securities market has not completely copied, so the exploration and based on the reference, to find suitable pricing methods in China stock market, to find a reasonable stock price has become the core issue in this paper. This paper studies the traditional pricing method, and compare the advantages and disadvantages between them. On this basis, the theory and method of looking for breakthrough, research found that the traditional method is not suitable for the existing Chinese motherboard market, and option, risk, neural network, introduce the game concept to IPO pricing a progress space for the motherboard market pricing, BP neural network method for it can reduce the error rate of about 15%, but we believe that accurate risk assessment and selection factors To improve the pricing accuracy, so we introduce the risk assessment and the factors were selected. At the same time, the differences of the listed environment is another cause of the pricing method cannot be directly applied mechanically, so we compared the representative of the United States, Singapore, Japan and Germany in the 4 countries, compared with the listing system I China. This paper explores method combined with BP neural network and VaR pricing risk for stock pricing, and that CVaR can measure the risk more accurately than the VaR method because the former conclusion, the error rate is low, the average reached only 12.69%, compared with the previous methods of BP neural network is reduced by more than 2%. At the same time to join the government: system factors the non quantifiable government system as listed cost factors, and using the normalization method of different dimension unity. In the study China, America, Singapore, Japan, the five countries of Germany Based on the market system, to further reduce the error rate to 11.96% by using the improved BP neural network model, the pricing is more effective, which is listed in the shortest time, the least error listed funds, pricing model of the German government supervision system under the strongest rate, at the same time, we come to the conclusion that supervision is time to market factors the most important of the three factors listed funds and supervision of the conclusions are of great help to the further reform of listed China government system.

【学位授予单位】：南京理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F832.51

【参考文献】