small positive and negative earnings that can create very large positive and
negative (often meaningless) P/Es. See Graham, Dodd, and Cottle (1962) for
long-run evidence supporting the low P/E approach and their mixed thoughts
on the price-to-book multiple.
- The BARRA growth factor is a predictor of future growth of a company and is
based on the five-year earnings-to-price ratio, historical earnings growth, re-
cent earnings change, recent I/B/E/S change, the current earnings-to-price ratio,
the I/B/E/S earnings-to-price ratio, and asset growth (BARRA, U.S. Equity
Beta Book, January 1996).
- The composite model-weighting scheme was advanced in Guerard (1990) and
continues to produce statistically significant rankings. It is obvious that an infi-
nite number of weighting schemes can be created; the four-period weighted re-
gression pattern produced significant real-time outperformance in the United
States and Japan during the 1988–1994 period. See Miller, Guerard, and
Takano (1992) and Guerard, Takano, and Yamane (1993).
- The Beaton-Tukey biweight procedure was put forth in Beaton and Tukey
(1974). The reader is referred to D. C. Montgomery and E. A. Peck, Introduc-
tion to Linear Regression Analysis(New York: John Wiley & Sons, 1982) for
a very complete description of the outlier-adjustment process.
- Guerard has experimented with several variations on equation (10.1) in his
joint research with Blin, Bender, Gultekin, Stone, Takano, and Yamane. Let us
briefly examine the average F-statistics and ICs of the various forms of equa-
tion (10.1) using the top 3,000 securities for the 1982–1994 period. In sum-
mary: (1) the BP variable has an average IC of 0.012 (t-value of 0.71), whereas
the EP variable has an average IC of 0.039 (t-value of 2.10), which indicates
that the low P/E, or high EP, strategy worked well in identifying undervalued
securities during the 1982–1994 period; (2) the use of relative variables—that
is, the relative EP (REP, the current EP divided by its five-year average of
monthly ratios)—increased the ICs of the four-fundamental-variable model
(EP, BP, CP, SP) from 0.039 (t-value of 2.17) to 0.042 (t-value of 2.28); (3) the
addition of the I/B/E/S FY1 forecast and breadth components further increased
the IC in (2) to 0.072 (t-value of 3.82); (4) the use of equation (10.1) in this
study produces an equally weighted IC of .058 (t-value of 3.15); and (5) the
Beaton-Tukey robust regression estimation procedure increased the ICs to ap-
proximately 0.085, with little difference in the composite model ICs. Guerard
initially used composite I/B/E/S revisions (CIR) and breadth (cm) in lieu of the
CTEF variable.
- It is interesting to note that if one uses only the 1,300-stock universe less the
socially screened stocks as the entire universe, reran the regression, and recal-
culated the expected returns, one finds an average F-statistic of 9.64 in the
OLS analysis and 12.4 in the ROB estimations. The average IC of 0.078 is sta-
tistically significant, having an average t-value of 3.63. The use of a value-
oriented model with the elimination of many smaller stocks does not diminish the
IC; however, the weighting of the composite growth variable is approximately
0.40. If one equally weights the seven-factor model, the average IC is 0.027
Notes 269
