McGraw-Hill

xvi
Foreword
each decision the RDM evaluates the expected utility of that decision
if each alternative.hypothesis were true. For a given decision,
the RDM computes a weighted sum of these expected utilities,
weighting each hypothesis by its probability. The (grand total) utility
the RDM attaches to the decision is this probability-weighted
sum. In particular, if one decision (for example, choice of portfolio)
would have high utility if the hypothesis that is considered most
likely were true, but would be disastrous if some different, not too
implausible, hypothesis were true, the decision’s (grand total) expected
utility would be less than that of a decision that would do almost
as well if the more likely hypothesis were true and not too
badly if the less likely hypothesis were true.
The human decision maker cannot perform a similar calculation,
at least not on an astronomically long of list alternative models of the
world. By imposing constraints that rule out extreme solutions, like
too large bets on particular securities, the HDM be may seen as intuitivelyemulatingthe
RDM by avoidingactionswith dire cons+
quences under not-too-implausible scenarios and hypotheses.
3. Chapter 13 of Markowitz (1959) discusses a many-period
consumption-investment game assuming perfectly liquid assets.
Lip service is given to the illiquid case, but only to recognize that
problem is importantandhard. In practice (for example,with
DPOS), transaction costs, including estimated market impact, and
constraints, such as upper bounds on portfolio tumover and the on
increase or decrease in holdings of a security at any one time, attempt
to achieve reasonable, if not optimal, policies in light of
illiquidity.
The inability of human decision makers to fully emulate RDMs
in maximizing expected utility in the face of uncertainty and
illiquidity is a manifestation of what Herbert Simon (1997) calls
“bounded rationality.” The imposition of more than minimally required
constraints, however, is not an example of what Simon calls
”satisficing” behavior. The investor does not add constraints that
lower ex ante efficiency because the investor is “satisfied” with less
efficiency. Constraints are added (in part, at least) because the investor
seeks protection against contingencies whose probability of
”disutility” is underrated by mean-variance approximation or, possibly,
by the parameter estimation procedure. We may view such
constraints as an effort by the HDM to achieve intuitively a policy