Bounded Rationality in Macroeconomics.
Vaughan, Mark D.
When former communists denounce Lenin, it makes headlines. A public
disavowal of rational expectations by Thomas Sargent would also make
headlines. In his new book, Bounded Rationality in Macroeconomics,
Sargent does not quite apostatize. Yet, he does concede that rational
expectations modeling leaves important research and policy questions
unanswered. To search for potential answers, Sargent surveys the
literature on bounded rationality. He also advocates bounded rationality
as a new approach to macroeconomic modeling. Though Sargent offers an
effective literature survey and several interesting applications of
bounded rationality, he fails to persuade the general reader to embrace
the new approach. Indeed, Sargent's reliance on mathematics at the
expense of intuition limits the audience of Bounded Rationality in
Macroeconomics to economists with highly technical research tastes.
Sargent's literature survey is effective because it clearly
contrasts modeling with bounded rationality and modeling with rational
expectations. In a rational expectations model, agents optimize subject
to constraints and hold mutually consistent views about the constraints.
Mutual consistency of views implies that agents know a great deal about
the environment. As Sargent puts it:
When implemented numerically or econometrically, rational
expectations models impute much more knowledge to the agents within the
model (who use the equilibrium probability distributions in evaluating
their Euler equations) than is possessed by an econometrician, who faces
estimation and inference problems that the agents in the model have
somehow solved [p. 3].
In contrast, bounded rationality drops the assumption of mutually
consistent perceptions. Dropping mutual consistency forces agents to act
like econometricians; now agents in the model must use theory and
statistics to learn about the environment. While more realistic as a
working assumption than rational expectations, bounded rationality
exacts a computational price. Again, in Sargent's words:
Ironically, when we economists make the people in our models more
"bounded" in their rationality and more diverse in their
understanding of the environment, we must be smarter, because our models
become larger and more demanding mathematically and econometrically [p.
2].
Researchers will pay the computational price if bounded rationality
promises a large return. Sargent sees large returns in three areas of
research: multiple equilibria, regime changes, and transitional
dynamics. One problem with rational expectations models is that they
generate multiple equilibria; bounded rationality can help identify a
subset of admissible equilibria. Another problem with rational
expectations models arises when analyzing regime changes. Analyzing
regime changes under rational expectations implies that agents
immediately accept the permanence of any change in policy rules;
analyzing regime changes under bounded rationality allows agents to
learn slowly about the consequences of policy changes. A third problem
is that rational expectations models can predict outcomes inconsistent
with real-world observations; bounded rationality offers new sources of
dynamics that better capture some real-world processes.
Sargent uses Jean Tirole's No-Trade theorem to demonstrate that
modeling with bounded rationality captures real-world processes
effectively. Under rational expectations, Tirole proved that equilibrium
prices in financial markets completely reveal private information
without trade. The zero-trading-volume implication of the theorem
strains credulity, given the observed volume in real-world financial
markets. Sargent shows that replacing Tirole's rational agents with
adaptive agents generates frictions capable of explaining volume.
Indeed, in simulations security prices with least squares learning
diverge from rational expectations prices, thereby generating volume,
for hundreds of periods.
The no-trade example ranks as Sargent's most interesting
application of bounded rationality. Appreciating the no-trade
application is difficult, however, because the exposition leans heavily
on mathematics at the expense of intuition. Consider an example lifted
from the section:
Since [z.sub.jt] is a subvector of [z.sub.t], system (24) can be used
to deduce the projections
E([z.sub.jt] [where] [Z.sub.jt-1]) = [S.sub.j]([Beta])[Z.sub.jt-1],
where [S.sub.j]([Beta]) depends on (T([Beta]), V([Beta])) and the
moments Eutu't. Thus, we have a mapping from a pair of perceived
laws of motion [Beta] = ([[Beta].sub.a], [[Beta].sub.b]) to a pair of
matrices ([S.sub.a]([Beta]), [S.sub.b]([Beta])) that determine optimal
(linear least squares) predictors. A rational expectations equilibrium
is a fixed point of this mapping [p. 117]. To be fair, a random passage
plucked from any mathematical treatise could bewilder the casual reader.
Still, the passage above proves difficult after reading the previous 116
pages carefully.
The entire middle section of Bounded Rationality proves equally
difficult. In Chapter four, for example, Sargent divides the
subject-networks and artificial intelligence - into seven daunting subheadings: the Perceptron, Feedforward Neural Networks with Hidden
Units, Associative Memory, Stochastic Networks, Local and Global
Methods, The Genetic Algorithm, and Classifier Systems. In the
perceptron discussion, Sargent confronts the reader with Heaviside step
functions and sigmoid functions. Later, in the associate memory
discussion, the reader trips over Hopfield networks and Hebb's
rule. The material under the other subheadings proves no easier. The
difficulty of the exposition is puzzling since Sargent seems to want a
wider audience than simply highly technical economists. Indeed, in the
closing pages of the book, he complains that ". . . macroeconomists
have shown very little interest in applying models of bounded
rationality to data." To reach a wider audience, Sargent could have
expanded the book, including enough intuition to allow macroeconomists
of all stripes to grasp the argument. Instead, he choose to argue at a
level that only a subset of macroeconomists can follow.
In Bounded Rationality in Macroeconomics, Thomas Sargent seeks to
inform the reader about bounded rationality and, more importantly, to
persuade him that bounded rationality is a valuable approach to
macroeconomic problems. The bounded rationality approach, Sargent
argues, makes agents in macroeconomic models behave more like
econometricians. Grasping the argument in Bounded Rationality requires,
unfortunately, that readers behave more like econometricians. The book
reads more like Macroeconomic Theory, a compilation of Sargent's
Ph.D. lecture notes, than Rational Expectations and Inflation, a
collection of his policy essays. In the introduction to Rational
Expectations and Inflation, Sargent observes:
One consequence of the highly technical orientation of early work on
rational expectations in macroeconomics is that an appreciation has been
slow to develop for the relevance of the ideas for the practice of
day-to-day macroeconomics [p. ix].
If Sargent had recalled the observation, he would have produced a
valuable introduction to bounded rationality modeling. But, by
emphasizing mathematics over intuition, Sargent reduced the value of
Bounded Rationality in Macroeconomics. Some future economist will
complain that macroeconomists failed to appreciate the relevance of
bounded rationality to practical problems. To express the complaint, the
economist can use Sargent's observation, with "bounded
rationality" in place of "rational expectations."
Mark D. Vaughan The Federal Reserve Rank of St. Louis