中文摘要：

鉴于CD统计量对截面相关形式是不稳健的,需对CD统计量进行修正。从而本文提出了渐近服从标准正态分布的MCD统计量,与已有的LMBC统计量和CD统计量比较发现：（1）三个统计量不存在明显的水平扭曲;（2）当截面间相关度有正有负,且规模较为接近时,CD统计量的功效很低,MCD统计量和LMBC统计量对于截面相关方式更稳健,而在同向截面相关模型中CD统计量的功效明显更好;（3）当模型存在分布误设或异方差时,CD统计量、LMBC统计量和MCD统计量的表现都较为稳健。我们将这三个统计量应用到国内上市公司现金分红行为模型中,发现公司间分红行为具有显著的相关性。

英文摘要：

Pesaran＇s CD statistic should be adjusted due to its instability for some kind of crosssectional dependence.This article proposes a new statistic named MCD statistic to test cross-sectional dependence,which is asymptotically distributed as N（0,1）.Monte Carlo simulation shows that：（1） CD statistic,Baltige et.al.＇s LMbc statistic and MCD statistic present no size distortion.（2） The power of CD statistic is very low when the cross-sectional correlated coefficients are eliminated mutually,while LMbc statistic and MCD statistic are robust to this elimination.But CD statistic exhibits better power in other situations.（3） CD statistic,LMbc statistic and MCD statistic are all robust to distribution misspecification and heteroscedasticity.We apply the statistics to test dependence of cash dividends payment behaviors of domestic listed companies,and find that the behaviors show a significant degree of cross-sectional dependence.