摘要：
In the Data Streaming world, screening for outliers is an often
overlooked aspect of the data preparation phase, which is needed
to rationalize inferences drawn from the analysis of data. In this
paper, we examine the effects of three outlier screens: A Trimming
Window, The Box-Plot Screen and the Mahalanobis Screen
on the market performance profile of firms traded on the NASDAQ
and NYSE. From among seven screening combinations tested, we identify
a single screening protocol that is the sequential application of
all three screens. This protocol is: (1) simple to program, (2)
significantly effective statistically and (3) does not compromise
power. This important result demonstrates that for the usual data
used by Financial Analysts there is one screening protocol that
can be relied upon to satisfy the outlier assumption of the regression
model used in generating the usual firm CAPM Return and Risk profile.
