http://clevelandfed.org/research/workpaper
Best available copy
-5-
the portfolio return, Λpt, and the measurement error in the return of the
market portfolio, A,ll scaled by the regression slope coefficient, β p.3
(3)
Γpt - Rft = a, + βp(rmt - Rft> + ʌpt - βpΛmt + ɛpt
Under the classical conditions, E<Λmt> = E(Apt) = O and Cov(Rpt,Λmt) =
Cov(Rpt,Λpt) = Cov(RmtjArnt) = Cov(RmtjApt) = Cov(AptjAmt) = O, nonzero Apt
causes the estimate of a, to be a high-biased estimate of the true a,
but it does not affect the estimate of β. Unfortunately, because βΛmt is
correlated with r,, the measurement error in the market portfolio causes
the estimate of βp to be low-biased. However, if one or more of the clas-
sical conditions fail to hold, the direction of the bias in the estimates of
a and β is generally ambiguous (see Maddala [20] c chapter 13).
During periods not characterized by flow-supply or flow-demand pressures,
the classical conditions should hold. Indeed, we argue that in the absence of
flow-supply and flow-demand pressures, recorded-price errors are random enough
across securities that Λmt is insignificant. Therefore, the remaining
source of bias in the estimated coefficients of equation (3) is Apt
<λ∣t for individual stock returns), which only affects estimates of a.
During periods of flow-supply or flow-demand pressures, both A and
Apt will be sources of bias in regressions on the market model. In addi-
tion, the flow-supply or flow-demand pressures will cause Amt and Apt
to be positively correlated and the estimate of β to be a high-biased estimate
of the true β.4 β estimates are high-biased because the positive
correlation between Amt and A,, causes the observed returns rpt and
r to be more highly correlated than the true returns Rpt and Rmt∙5