基于小波神经网络的高阶CAPM实证研究  

Empirical Study on High Moment CAPM Based on Wavelet Neural Network

作　　者：戴念念[1] 陈小伟[1] 

机构地区：[1]中南财经政法大学金融学院

出　　处：《投资研究》2011年第12期96-111,共16页Review of Investment Studies

摘　　要：本文在传统CAPM的基础上,引入了一个高阶的CAPM。借助小波神经网络在非线性函数逼近方面的优势,使用上海证券交易所股票数据分别对二阶至四阶CAPM进行了实证分析。最终的研究结果表明:就上海股市而言,12只大盘股组合已经能够有效分散非系统风险,而12只小盘股不能充分化解非系统风险,存在所谓的"规模效应";训练后的网络预测显示,高阶CAPM无论是在预测精度还是预测稳定性上都要明显优于传统的CAPM,在一个非系统风险得到充分分散的证券组合中,加入三阶矩的CAPM已经能够比较准确地把握风险资产的市场定价。In this paper, we introduce a high moment Capital Asset Pricing Model （CAPM）. Using the method of Wavelet Neural Network to fit nonlinear CAPM, empirical studies have been conducted from second moment to fourth moment respectively with daily stock exchange data of Shanghai market. The final result indicates that： unsystematic risk could be dispersed effi- ciently with a portfolio including twelve large cap stocks. But that can not be reached with a portfolio of twelve small cap stocks. This means there is a scale effect among small cap stocks. The network forecast after training shows that： CAPM with high moment is better than low moment both in prediction accuracy and stability. And in a perfect security portfolio, asset price could be forecasted closely by a third moment CAPM.

关 键 词：小波网络 高阶矩 CAPM 

分 类 号：F224[经济管理—国民经济] F832.51