基于K均值聚类和主成分分析上市公司业绩评价指标研究.docVIP

基于K均值聚类和主成分分析的上市公司业绩评价指标研究 JIANG Bing，WANG Qian，DING Hu-song School of Management, Hefei University Of Technology, P.R.China,230009摘要：评价指标体系的确定是上市公司经营业绩评价研究的一个重要方面，其研究的立足点是如何权衡指标的全面性和代表性。本文提出了综合运用多元统计分析中的K均值聚类分析和主成分分析对上市公司众多业绩指标进行先分类再筛选的指标确定方法。首先运用K均值聚类分析对备选指标进行分类，并根据F统计值的大小，确定适宜的分类数；然后对含有较多指标的类运用主成分分析进一步筛选指标。在实证研究中，以车类板块28家上市公司自2003-2005连续三年的32个业绩指标为数据样本，运用K均值聚类分析和主成分分析，结合实际意义，最终从32个备选指标中选出12个指标；通过将12个和32个指标所得出的指标归一化平均值所作的评价进行比较，表明筛选的12个指标具有很好的全面性和代表性。 关键词：F统计值，指标选择，K均值聚类，上市公司，主成分分析 Abstract: Establishment of index systems evaluating listed companies business performance is an important aspect of company performance evaluation which is based on how to weigh the entirety and typification of indices systems. In this paper,a method integrating K-means clustering with principal component analysis in multivariant statistics to classify and select indices for listed companies performance evaluation is proposed. first the classification is decide using K-means clustering and the number of cluster is determined according to F-statistic. Then the indices in the classes including more indices are filtrated by principal component analysis.b In practical study,taking 32 performance indices of 28 stocks in vehicle trade from 2003 to 2005 as example,12 indices are chosen from 32 indices to be selected at last by integrating K-clustering with principal component analysis. Comparing the evaluation according to the average of unitary indices calculated by 12 indices with that calculated by 32 indices verifies that 12 indices selected are of good entirety and typification. Keywords: F-statistic, Index selection, K-means clustering, Listed companies, Principal component analysis 1 引言 随着我国证券市场的发展，对上市公司经营业绩的综合评价已成为理论与实践界研究的热点。其中关于综合评价指标体系的建立，目前的做法一般是从上市公司披露的数据指标中选择一部分构成指标体系，且多数是出于评价主体需要而人为设定[1][2]。众所周知，对综合评价指标的选择，既要求指标能很好地反映研究对象某方面的特性，即代表性；又要求能反映对象的全部信息，即全面性[3]。若要满足全面性, 势必要增加指标数量，而增加了指标数量，指标间相关程度可能增大，又影响了代表性。对上市公司来说，媒体披露的是大量的指标数据序列，仅财务指标就多达56项。如何从众多的数据信息中提炼出满足评价要求的少数指标，尤其是如何权衡指标的全面