基于案例推理的股价操纵行为判别系统研究

摘要

自中国股票交易市场成立以来,各种各样的股价操纵行为层出不穷.就如何对股价操纵行为进行判别做了研究,设计了一个基于案例推理的股价操纵行为判别系统.利用1996～2009年间在中国股市中被公开披露和查处的股价操纵案构建案例库,在前人研究的基础上从价量和风险两个角度提取了股票的若干市场交易属性作为判别指标,并通过二值Logistic回归分析确立了指标属性的权重;随后采用K最近邻法作为案例检索算法,依据指标属性的特征同时确立了属性相似度的计算方法,并利用SQL语言对核心算法在数据库中进行了程序实现.同时,通过设计"阈值控制法"来维护案例库,确保了案例库中知识的代表性,并能有效地降低信息冗余度.基于案例推理的股价操纵行为判别系统能够绕过"知识获取"这个难题,克服基于规则的推理系统的一些弱点,为今后股价操纵行为的判别研究提供了一个新的方向.%Since the establishment of the stock exchange market of China, there are lots of market manipulating behaviors. On how to detect stock price manipulations and designed a case-based reasoning system is studied for manipulation detection. Published stock price manipulation cases of China' s stock exchange market from 1996 to 2009 is used to construct the case database, extracted some key factors which is related to the stock price manipulation detection from the stock trading features based on the previous researches, and used Logistic regression method to identify the weight of each key factors in the detection. Then K-nearest neighbor algorithm is used to retrieve cases, established the likelihood calculation algorithms for the extracted key factors, and used SQL language to program all the key algorithms in database. Meanwhile, a threshold controlling mechanism is designed to maintain the database in order to ensure the representativity of the cases stored in case database and reduce the information redundancy. The case-based reasoning system for detection of stock price manipulation jumps over the problem of "knowledge acquisition", overcomes the weaknesses of rule-based reasoning, and provides a new direction for researches on detection of stock price manipulation.