Formulating and Implementing Investment Strategies Using Financial Econometrics 307
trading strategies. The process involves two phases. The first is quantita-
tive research utilizing financial econometrics. The second involves using
the results obtained from a financial econometrics study to develop an
investment strategy.
The Quantitative Research Process
As can be seen in Figure 15.1, the financial econometrics research aspect of
the quantitative research process involves three phases:
- Develop an ex ante justification based on financial economic theory.
- Select a survivorship-free sample.
- Select a methodology to estimate the model.
We discuss each phase in this section.Develop an ex ante Justification Based on Financial
economic theory
A sound hypothesis is a necessary condition for the successful formulation
of an implementable and replicable investment strategy. Financial econom-
ics, however, can only be motivated with creative intuitions and scrutinized
by strict logical reasoning, but it does not come from hindsight or prior
experience. This requirement is critical since scientific conclusions can easily
be contaminated by the process of data snooping, especially when a truly
independent financial economic theory is not established first.
As explained in the previous chapter, data snooping is identifying seem-
ingly significant but in fact spurious patterns in the data.^1 All empirical tests
are at risk for this problem, especially if a large number of studies have been
performed on the same data sets. Given enough time and trials, people who
are convinced of the existence of a pattern will eventually manage to find
that pattern, real or imagined. Furthermore, there is an identical life cycle of
experience in data snooping. Researchers are often confronted with exactly
the same issues and will have to make the same types of choices in the process.
The process of data snooping comes in several forms. At some basic but
subtle level, a hypothesis based on financial economics is founded by the
knowledge of past patterns in data. Researchers may establish their “prior”
(^1) See, for example, Stephen A. Ross, “Survivorship Bias in Performance Studies,” and
Andrew Lo, “Data-Snooping Biases in Financial Analysis,” both appearing in Pro-
ceedings of Blending Quantitative and Traditional Equity Analysis (Charlottesville,
VA: Association for Investment Management and Research, 1994).