400
The Nobel Prizes
This essay proceeds as follows. In Section 2, I review the development of
time series econometric modeling, including the initiation of rational expecta-
tions econometrics. In Section 3, I review my contributions to the economet-
ric study of partially specified models, adapting to the study asset pricing and
macroeconomic uncertainty. I describe methods and approaches to the study of
fully specified models based on asset pricing considerations in Section 4. In Sec-
tion 5, I explore the consequences for asset pricing models when investor beliefs
are not in full accord with an underlying model, which can result in investor be-
havior that resembles extreme risk aversion. In Section 6, I review perspectives
on model ambiguity which draw on work by decision theorists and statisticians
to revisit the framework that I sketch in Section 5. I draw some conclusions in
Section 7.
2 RaTioNal exPeCTaTioNS eCoNoMeTRiCS
Rational expectations econometrics explores structural stochastic models of
macroeconomic time series with the ambition to be a usable tool for policy
analysis. It emerged in response to a rich history of modeling and statistical
advances. Yule (1927) and Slutsky (1927, 1937) provided early characterizations
of how time series models can generate interesting cyclical behavior by propa-
gating shocks. Yule (1927) showed that a second-order autoregression could
reproduce intriguing patterns in the time series. He fit this model to sunspot
data, known to be approximately but not exactly periodic. The model was built
using independent and identically distributed (iid) shocks as building blocks.
The model produced a damped periodic response to random impulses. Simi-
larly, Slutsky (1927, 1937) constructed models that were moving-averages of iid
shocks and showed how such processes could be arbitrarily close to exact peri-
odic sequences.
2
He also demonstrated how moving-average type models could
account for British business cycle data.
Frisch (1933), who shared the first Sveriges Riksbank Prize in Economics
with Tinbergen, pushed this agenda further by exploring how to capture dy-
namic economic phenomenon through probability models with explicit eco-
nomic underpinnings. Frisch discussed propagation from initial conditions
and described an important role for random impulses building in part on the
work of Yule (1927) and Slutsky (1927, 1937). In effect, Frisch (1933) introduced
2
I cite two versions of Slutsky’s paper. The first one was published in Russian. The second
one was published in English a decade later with a more comprehensive set of results.
English translations of the first paper were circulated well in advance of 1937.
6490_Book.indb 400
11/4/14 2:30 PM
Uncertainty Outside and Inside Economic Models 401
impulse response functions to economics as a device to understand the inter-
temporal impact of shocks on economic variables. Haavelmo (1944) took an
additional step by providing foundations for the use of statistical methods to
assess formally the stochastic models. This literature set the foundation for a
modern time series econometrics that uses economics to interpret evidence in a
mathematically formal way. It featured important interactions among econom-
ics, mathematics and statistics and placed a premium on formal model build-
ing.
3
Haavelmo (1944) confronts uncertainty as an econometrician outside the
model that is to be estimated and tested.
Investment and other decisions are in part based on people’s views of the
future. Once economic decision makers are included into formal dynamic eco-
nomic models, their expectations come into play and become an important in-
gredient to the model. This challenge was well appreciated by economists such
as Pigou, Keynes and Hicks, and their suggestions have had durable impact on
model building. Thus, the time series econometrics research agenda had to take
a stand on how people inside the model made forecasts. Alternative approaches
were suggested including static expectations, adaptive expectations or appeals
to data on beliefs; but these approaches left open how to proceed when using
dynamic economic models to assess hypothetical policy interventions.
A productive approach to this modeling challenge has been to add the hy-
pothesis of rational expectations. This hypothesis appeals to long histories of
data to motivate the modeling expectations. The Law of Large Numbers gives an
approximation whereby parameters that are invariant over time are revealed by
data, and this revelation gives a model builder a way to formalize the expecta-
tions of economic investors inside our models.
4
This approach to completing the
specification of a stochastic equilibrium model was initiated within macroeco-
3
Frisch, in particular, nurtured this ambitious research agenda by his central role in the
foundational years of the Econometric Society. His ambition is reflected in the 1933 mis-
sion statement he wrote for the journal Econometrica: “. . . Experience has shown that
each of these three viewpoints, that of statistics, economic theory, and mathematics, is a
necessary, but not by itself a sufficient, condition for a real understanding of the quantita-
tive relations in modern economic life. It is the unification of all three that is powerful.
And it is this unification that constitutes econometrics.” Frisch (1933b).
4
More than three hundred years ago, Jacob Bernoulli proved a result that implied a Law
of Large Numbers. He was motivated in part by social problems for which probabilities
had to be estimated empirically, in contrast to typical gambling problems. Bernoulli’s
result initiated an enduring discussion of both the relevance of his simple model speci-
fication and of the approximation he established. See Stigler (2014) for an interesting
retrospective on Bernoulli’s contribution.
6490_Book.indb 401
11/4/14 2:30 PM