variables, such as reported earnings, book value, cash flow, and sales. The
composite earnings valuation model is statistically significant, but not as
effective in asset selection as the composite earnings forecasting variable by
itself during the 1990–2001 period. Corporations seek to enhance stock-
holder wealth by paying dividends and engaging in capital expenditures
and research and development. We find that stockholder wealth is in-
creased by including research and development expenditures with the com-
posite earnings forecasting variable. Risk models have been constructed to
analyze the covariance matrix of U.S. security returns in terms of market
risk, the security beta, and extramarket covariance.
The factor loading of the variable is estimated and analyzed in the uni-
verses. The use of an R&D quadratic variable enhances stockholder re-
turns and wealth relative to the use of a stock valuation model, and we
address the issue of the factor loading of the R&D variable.
Efficient Portfolio Optimization Results
The universes for this study are the monthly Frank Russell stock universes
for the January 1990–December 2001 period. The information coefficient
(IC) analysis introduced in Chapter 8 supported the construction of a com-
posite earnings forecast model in that the IC of the composite model ex-
ceeded the ICs of its components. We address in this section the estimated
asset selection properties of the earnings components and the composite
models. Let us address the estimated earnings forecasting components of
the CTEF model for the Russell 3000 universe during the 1990–2001 pe-
riod. The CTEF model produced not only higher ICs than its components,
but also higher and more statistically significant asset selection than its
components in the Russell 3000 universe, as we saw in Chapter 8. The
CTEF variable produces statistically significant total active returns and as-
set selection (see Table 9.1). We test the Frank Russell large market capital-
ization universe (the Russell 1000), middle market capitalization
(mid-cap), small capitalization (Russell 2000), and small and middle mar-
ket capitalization (Russell 2500) universes. The CTEF produces statisti-
cally significant active returns in all Frank Russell universes, although the
returns rise substantially as the size of the firms decrease, and we move into
the Russell 2000 securities. We test the equally weighted composite model,
CTEF, of I/B/E/S earnings forecasts, revisions, and breadth, described in the
previous section. The portfolio optimization algorithm seeks to maximize
the ranking of the CTEF variable while minimizing risk.
The underlying CTEF variable is statistically significant, having a
monthly information coefficient of 0.049 over the 491,119 observations.