Formulating and Implementing Investment Strategies Using Financial Econometrics 313
magnitude of the event;^9 stocks are purchased for their glamour but not
for intrinsic value;^10 and, low price-earnings stocks paying high returns do
not imply that high price–earnings stocks pay low returns.^11 We are not
proposing that a statistical model should include all these phenomena, but
the modeling methodology should be flexible enough to entertain such pos-
sibilities if they are warranted by the theory.
Statistical Significance Does Not guarantee alpha
Staunch defenders of quantitative research argue that profitable investment
strategies cannot be commercialized by quantitative analysis using tools
such as those available to financial econometricians;^12 the production of
excess returns will stay idiosyncratic and proprietary. Alpha will originate
in those proprietary algorithms that outperform commercially standardized
packages for data analysis. In other words, researchers will have to learn to
gain confidence even if there is no statistical significance, while statistical
significance does not guarantee alpha.
Since quantitative market strategists often start with the identification of
a pattern that is defined by financial econometric tools, it is easy to assume
alpha from conventional statistical significance. To show that there is not nec-
essarily a link, we perform a typical momentum trading strategy that is solely
based on the predictability of future returns from past returns. A simplified
version of the return-generating process under this framework follows:
Et–1(Rt) = a + bt–1Rt–1
where Et–1(Rt) is the conditional expected return for the period t, evaluated
at point t – 1, a is time-invariant return, and bt–1 is the momentum coefficient
observed at time t – 1. When bt–1 is (statistically) significantly positive, the
time-series returns are said to exhibit “persistence and positive momentum.”
To implement the trading strategy using the information in correlations,
stocks with at least a certain level of correlation are included in portfolios
at the beginning of each month, and their returns are tracked. The perfor-
mance of these portfolios apparently reflects the statistical significance (or
lack thereof) in correlation between successive returns. In Table 15.3, we
(^9) See Christopher K. Ma, “Preference Reversal in Futures Markets,” working paper,
Stetson University, 2010.
(^10) See Josef Lakonishok, Andrei Shleifer, and Robert W. Vishny, “Contrarian Invest-
ment, Extrapolation, and Risk,” Journal of Finance 49, no. 5 (1994): 1541–1578.
(^11) See Ma, “How Many Factors Do You Need?”
(^12) See, for example, Russell H. Fogler, “Investment Analysis and New Quantitative
Tools,” Journal of Portfolio Management (1995): 39–47.