402
The Nobel Prizes
nomics by Muth (1961) and Lucas (1972). Following Lucas (1972) in particular,
rational expectations became an integral part of an equilibrium for a stochastic
economic model.
The aim of structural econometrics is to provide a framework for policy
analysis and the study of counterfactuals. This vision is described in Marschak
(1953) and articulated formally in the work of Hurwicz (1962). While there are
a multitude of interesting implications of the rational expectations hypothesis,
perhaps the most important one is its role in policy analysis. It gives a way to
explore policy experiments or hypothetical changes that are not predicated on
systematically fooling people. See Sargent and Wallace (1975) and Lucas (1976)
for a discussion.
5
From an econometric standpoint, rational expectations introduced impor-
tant cross-equation restrictions. These recognize that parameters governing
the dynamic evolution of exogenous impulses to the model must also be pres-
ent in decision rules and equilibrium relations. These restrictions reflect how
decision-makers within the model are forward-looking. For instance, an invest-
ment choice today depends on the beliefs about how profitable such investments
will be in the future. Investors forecast the future, and the rational expectations
hypothesis predicts how they do this. The resulting cross-equation restrictions
add a new dimension to econometric analysis; but these restrictions are built
on the premise that investors have figured much out about how the future will
evolve. See Sargent (1973), Wallis (1980) and my first published paper, Hansen
and Sargent (1980), for characterizations of these restrictions.
6
To implement
this approach to rational expectations econometrics, a researcher is compelled
to specify correctly the information sets of economic actors.
7
When building
actual stochastic models, however, it is often not clear what information should
be presumed on the part of economic agents, how they should use it, and how
much confidence they have in that use.
The introduction of random shocks as impulses to a dynamic economic
model in conjunction with the assumption of rational expectations is an exam-
ple of uncertainty inside a model. Under a rational expectations equilibrium, an
5
To be clear, rational expectations offers an approach for comparing distinct stochastic
equilibria but not the transitions from one to another. For an interesting extension that
allows for clustering of observations near alternative self-confirming equilibria in con-
junction with escapes from such clusters see Sargent (1999).
6
While this was my first publication of a full length paper, this was not my first publica-
tion. My first was a note published in Economic Letters.
7
See Sims (2012) for a discussion of the successes and limitations of implementing the
Haavelmo (1944) agenda to the study of monetary policy under rational expectations.
6490_Book.indb 402
11/4/14 2:30 PM
Uncertainty Outside and Inside Economic Models 403
investor inside the model knows the model-implied stochastic evolution for the
state variables relevant for decision making and hence the likely consequences
of the impulses. An econometrician also confronts uncertainty outside a model
because of his or her lack of knowledge of parameters or maybe even a lack of
confidence with the full model specification. There is an asymmetry between
the inside and the outside perspectives found in rational expectations econo-
metrics that I will turn to later. But first, I will discuss an alternative approach to
imposing rational expectations in econometric analyses.
3 RobuST eCoNoMeTRiCS uNdeR RaTioNal exPeCTaTioNS
My econometrics paper, Hansen (1982b), builds on a long tradition in econo-
metrics of “doing something without having to do everything.” This entails the
study of partially specified models—that is, models in which only a subset of
economic relations are formally delineated. I added to this literature by ana-
lyzing such estimation problems in greater generality, giving researchers more
flexibility in modeling the underlying time series while incorporating some ex-
plicit economic structure. I studied formally a family of Generalized Method of
Moments (GMM) estimators, and I adapted these methods to applications that
study linkages between financial markets and the macroeconomy.
8
By allow-
ing
for partial specification, these methods gain a form of robustness. They are
immune to mistakes in how one might fill out the complete specification of the
underlying economic model.
The approach is best thought of as providing initial steps in building a time
series econometric model without specifying the full econometric model. Con-
sider a research program that studies the linkages between the macroeconomy
and financial markets. One possibility is to construct a fully specified model of
8
My exposure to using GMM estimators as a vehicle to represent a broad family of esti-
mators originally came from Christopher Sims’ lectures. As
a graduate student I became
interested in central limit approximations that allow for econometric error terms to pos-
sess general types of temporal dependence by using central limit approximations of the
type demonstrated by Gordin (1969). I subsequently established formally large sample
properties for GMM estimators in such circumstances. Interestingly, Econometrica chose
not to publish many of the formal proofs for results in my paper. Instead they were pub-
lished thirty years later by the Journal of Econometrics, see Hansen (2012). Included in
my original submission and in the published proofs is a Uniform Law of Large Numbers
for stationary ergodic processes. See Hansen (2001) and Ghysels and Hall (2002) for
further elaborations and discussion about the connection between GMM and related
statistics literatures. See Arellano (2003) for a discussion of applications to panel data.
6490_Book.indb 403
11/4/14 2:30 PM