McGraw-Hill

On the Value of "Value" 127
22. On low P/E, see Shiller (1984). For a practitioner's viewpoint, see
Dreman (1982).
23. On splits, see Ohlson and Penman (1985); on earnings, see DeBondt
and Thaler (1987); on events, see Renshaw (1984), Howe (1986), and
Brown and Harlow (1988).
24. This bias is also consistent with Prospect Theory's base-rate fallacy.
25; For an extensive discussion of these attributes and our methodology,
see Jacobs and Levy (1988~).
26. For a description of the threestage DDM, see Sharpe (1985), Chapter
14.
27. The universe each quarter was a subset of the 1500 largestcapitalization
stocks for which all the necessary data were available
to calculate DDM expected returns. The sample size ranged from a
low of 1035 to a high of 1337.
28. The correlation, significant at the 1 percent level, between the
unanticipated component of consecutive earnings
announcements indicates that surprises tend to repeat. This may
be due to the behavior of analysts, if they do not fully
incorporate all relevant information in forming their earnings
expectations. This persistence of earnings surprise is consistent
with Jones, Rendleman, and Latane (1984). Rendleman, Jones,
and Latane (1986) propose that most of the postannouncement
stock response to an earnings surprise may actually be a
preannouncement adjustment to next quarter's surprise.
29. See Michaud and Davis (1982) and Bethke and Boyd (1983) for
earlier documentation on growth, P/E, and yield.
30. For discussion of this analyst bias, see Elton, Gruber, and Gultekin
(1984). Michaud and Davis (1982) document a negative correlation of
DDM expected return with long-term growth. Our earnings torpedo
measure is not comparable because it looks ahead only 1 fiscal year.
31. Note that we lagged price 1 month to avoid bid/ask and pricing
error problems for most measures, as discussed in Jacobs and Levy
(1988~). DDM and residual-return reversal are notable exceptions.
32. Alternatively, a relationship could be spuriously induced by a price
. change near quarter-end consistent with a value change that has not
yet been reflected in the DDM model. This may occur because of lags
in updating consensus earnings databases for revisions in analysts'
estimates Uacobs and Levy (1988c)l.
33. For an extensive discussion of calendar-based anomalies, see Jacobs
and Levy (1988a) and Ferson, Foerster, and Keim (1987).