7Heteoskedasticity.pptVIP

Topic 7: Heteroskedasticity Properties of OLS under Heteroscedasticity Test for Heteroscedasticity Weighted Least Square Estimator * 擦扬痔拐赖嘱猿太魔痪氮镶久韵涅辕沁拔京狸咱或飞榨侨沂仰尹飘觅奇蔽7Heteroskedasticity7Heteroskedasticity Heteroscedastic: variance are not constant across observations. The problems could arise in both the cross-section and time series data set. For example: Variation in profit tends to larger for the larger firms than the smaller ones. Greater variation on expenditure among high-income families than low ones In the heteroscedastic case, (still assume non-autocorrelation) * 检几返腿尚垦霄镍硫元程坚希缘软共琼颇瘁衣耸侠穆忿捎多养狂密掳家酗7Heteroskedasticity7Heteroskedasticity * This form is an arbitrary scaling, which allow us to use a normalization, wi can be thought of as the weights to reflect the variety in the disturbance variance. So, the homoscedastic disturbance is the special case with wi =1 In the case of heteroskedastic, from the last topic, OLS is unbiased, consistent, asymptotically normally distributed, but Inefficient. The big problem is that usual covariance estimator for OLS is not appropriate any more, the appropriate covariance matrix of OLS is, 淮札粳戌罢擂庶妒腋迁稍具菱游光悲甥簿脸佳倘屎逆芬垄鸡皱峰夜妻估趴7Heteroskedasticity7Heteroskedasticity *Given thatSo, the asymptotic distribution of OLS is 暮扳责稗乍镜萤夺拿按魁个峡卤奢号恿味滔熔稗呆钧谚打刁季找搏载款敖7Heteroskedasticity7Heteroskedasticity The estimated covariance estimator of b How difference between the conventionally estimated covariance of b and the appropriate covariance estimator? The usual covariance estimator of b is 宵接谨症闻告泳蓬勉酬钠沧蓝囤牵罕伸称山帘原带闯设譬脆畏掀枣氮温伶7Heteroskedasticity7Heteroskedasticity 担揣弓繁乎幽滑么瞧毕宰疗搬敏独涝棠熄持攀曾鲤救搓壶拿践姓倾灿杉谤7Heteroskedasticity7Heteroskedasticity Then We can also show that So, So, In large sample, the difference is approximately is 每属狡储屹渠蹲必沏筛肖硅贫门宁羡蕴兢投买邢匿绢浚释著疏此出姜藻尿7Heteroskedasticity7Heteroskedasticity Consider the simple case where X contains only one columns regressors. By the C.L.T.,By the construction of wi , So, If the