weighted model despite turnover of more than 170 percent. The regression-
weighted composite model has an average F-statistic of 28 and is statisti-
cally significant at the 5 percent level. The composite model expected
return ranking procedure described earlier produces an average informa-
tion coefficient of 0.093 for the 1982–1996 period (t = value of 6.5) and
an average t-value of 4.14 for the 1987–1996 period. The lower quintile
(least preferred) securities consistently underperform the average stock
return, and the upper quintile (most preferred) securities produce positive
excess returns such that the quintile spread is positive and statistically
significant. The information coefficient, measuring the association be-
tween the ranked composite model score and subsequent ranked total re-
turns, indicates that the quantitative model is statistically significant in its
ranking of securities. The IC is a standard tool used in accessing the pre-
dictive power of financial information (Farrell 1983). In a more recent
study, Guerard, Blin, and Bender (1996b) found that the estimated model
from equation (10.1) outperformed the S&P 500 index by 420 basis
points annually during the 1987–1994 period, assuming a 3 percent up-
per bound on security weights, transactions costs of 80 basis points each
way, and quarterly reoptimization. Guerard, Gultekin, and Stone (1996)
found excess returns of approximately 412 basis points annually during
the 1982–1994 period using a variation on equation (1) that was dis-
cussed in Chapter 8.^10 The Guerard, Blin, and Bender (1996a,b) and
Guerard, Gultekin, and Stone (1996) studies used unscreened investment
universes.
The estimated expected return ranking model is used to create portfo-
lios during the 1987–1995 period using a socially screened universe. The
socially screened universe is created by subtracting the current KLD exclu-
sions from a universe of 1,200 large stocks, resulting in a screened universe
of approximately 950 stocks. A simulation is run for the January
1987–December 1995 period on the socially screened universe in which
one tightly constrains industry and capitalization weighting and a 100-basis-
point transactions cost (round-trip) is assessed in the simulation. We find
that the estimate composite model produces an average excess return of
743 basis points. Socially screened portfolios can outperform unscreened
portfolios during the 1987–1995 period. One can invest in a socially
screened portfolio and still outperform the S&P 500 socially screened
benchmark.^11 It is interesting to see how the use of a socially screened uni-
verse creates a higher average weight of the proprietary growth variable in
equation (10.1). The ICs of the composite model may be enhanced as one
shifts from a more value-oriented weighting to a more growth-oriented
weighting as one forecasts relative factor returns.^12
backadmin
(backadmin)
#1