404
The Nobel Prizes
the macroeconomy including the linkages with financial markets that are pre-
sumed to exist. This is a lot to ask in early stages of model development. Of
course, an eventual aim is to produce a full model of stochastic equilibrium.
The econometric tools that I developed are well suited to study a rich fam-
ily of asset pricing models, among other things. Previously, Ross (1978) and
Harrison and Kreps (1979) produced mathematical characterizations of asset
pricing in frictionless asset pricing markets implied by the absence of arbitrage.
Their work provides a general way to capture how financial markets value risky
payoffs. My own research, and that with collaborators, built on this conceptual
approach, but with an important reframing. Our explicit consideration of sto-
chastic discounting, left implicit in the Ross (1978) and Harrison and Kreps
(1979) framework, opened the door to new ways to conduct empirical studies
of asset pricing models using GMM and related econometric methods. I now
describe these methods.
3.1 a GMM approach to empirical asset Pricing
A productive starting point in empirical asset pricing is
E
S
t
+
S
t
⎛
⎝⎜
⎞
⎠⎟
Y
t
+
⎪F
t
⎡
⎣
⎢
⎤
⎦
⎥ =
Q
t
(1)
where S > 0 is a stochastic discount factor (SDF) process. In formula (1), Y
t+l
is a
vector of payoffs on assets at time t + l, and Q
t
is a vector of corresponding asset
prices. The event collection (sigma algebra), F
t
, captures information available
to an investor at date t. The discount factor process is stochastic in order to
adjust market values for risk. Each realized state is discounted differently and
this differential discounting reflects investor compensation for risk exposure.
Rational expectations is imposed by presuming that the conditional expecta-
tion operator is consistent with the probability law that governs the actual data
generation. With this approach a researcher does not specify formally that prob-
ability law and instead “lets the data speak.”
Relations of type (1) are premised on investment decisions made in optimal
ways and are fundamental ingredients in stochastic economic models. The spec-
ification of a SDF process encapsulates some economics. It is constructed from
the intertemporal marginal rates of substitution of marginal investors. Investors
consider the choice of consuming today or investing to support opportunities to
consume in the future. There are a variety of investment opportunities with dif-
ferential exposure to risk. Investors’ risk aversion enters the SDF and influences
the nature of the investment that is undertaken. While I have used the language
of financial markets, this same formulation applies to investments in physical
6490_Book.indb 404
11/4/14 2:30 PM
Uncertainty Outside and Inside Economic Models 405
and human capital. In a model of a stochastic equilibrium, this type of relation
holds when evaluated at equilibrium outcomes. Relation (1) by itself is typi-
cally not sufficient to determine fully a stochastic equilibrium, so focusing on
this relation alone leads us to a partially specified model. Additional modeling
ingredients are required to complete the specification. The presumption is that
whatever those details might be, the observed time series come from a stochas-
tic equilibrium that is consistent with an equation of the form (1).
Implications of relation (1), including the role of SDFs and the impact of
conditioning information used by investors, were explored systematically in
Hansen and Richard (1987). But the origins of this empirically tractable for-
mulation traces back to Rubinstein (1976), Lucas (1978) and Grossman and
Shiller (1981), and the conceptual underpinnings to Ross (1978) and Harrison
and Kreps (1979).
9
To implement formula (1) as it stands, we need to specify the
information set of economic agents correctly. The Law of Iterated Expectations
allows us to understate the information available to economic agents.
10
For in-
stance let F
t
⊂ F
t
denote a smaller information set used by an external analyst.
By averaging over the finer information set F
t
conditioned on the coarser infor-
mation set F
t
, I obtain
E
S
t
+
S
t
⎛
⎝⎜
⎞
⎠⎟
Y
t
+
( )
′ −
Q
t
( )
′⎪F
t
⎡
⎣
⎢
⎤
⎦
⎥ = 0. (2)
I now slip in conditioning information through the “back door” by constructing
a conformable matrix Z
t
with entries in the reduced information set (that are
F
t
measurable). Then
E
S
t
+
S
t
⎛
⎝⎜
⎞
⎠⎟
(Y
t
+
′)Z
t
−(Q
t
′)Z
t
⎪F
t
⎡
⎣
⎢
⎤
⎦
⎥ = 0.
9
The concept of a SDF was first introduced in Hansen and Richard (1987). Stochastic
discount factors are closely connected to the “risk-neutral” probabilities used in valu-
ing derivative claims. This connection is evident by dividing the one-period SDF by its
conditional mean and using the resulting random variable to define a new one-period
conditional probability distribution, the risk neutral distribution.
10
In his study of interest rates, Shiller (1972) in his PhD dissertation
suggested omitted
information as a source of an “error term” for an econometrician. In Hansen and Sargent
(1980), we built on this insight by contrasting implications for a “Shiller error-term” as
a disturbance term to processes that are unobserved to an econometrician and enter
structural relations. In Hansen and Sargent (1991) we show how to allow for omitted
information in linear or log-linear time series models using quasi-likelihood methods.
6490_Book.indb 405
11/4/14 2:30 PM