418
The Nobel Prizes
the aversion of the decision maker to consumption risk. Applied researchers
have only been too happy to explore this channel. A fully solved out stochastic
equilibrium model represents C and V as part of the model solution. For in-
stance log C might have an evolution with the same form as log G as specified
in (9) along a balanced stochastic growth trajectory. Representing S as in (8)
presumes a solution for V
t
or more conveniently
V
t
C
t
as a function of X
t
along
with a risk adjusted counterpart to V
t
and these require a full specification of
investor information.
For early macro-finance applications highlighting the computation of
continuation values in equilibrium models, see Hansen et al. (1999) and Tal-
larini (2000). The subsequent work of Bansal and Yaron (2004) showed how
these preferences in conjunction with forward looking beliefs about stochastic
growth and volatility have a potentially important impact on even one-period
(in discrete time) or instantaneous (in continuous time) risk prices through the
forward-looking channel. Hansen (2011) and Borovička et al. (2011) show that
the prices of growth rate shocks are large for all payoff horizons with recursive
utility and when γ is much larger than ρ. By contrast, for power utility models
with large values of ρ = γ, the growth rate shock prices start off small and only
eventually become large as the payoff horizon increases. The analyses in Han-
sen et al. (2008) and Restoy and Weil (2011) also presume that one solves for
the continuation values of consumption plans or their equivalent. This general
approach to the use of recursive utility for investor preferences makes explicit
use of the information available to investors and hence does not allow for the
robustness that I discussed in section 3.
26
Sometimes there is a way around this sensitivity to the information structure
when conducting an econometric analysis. The empirical approach of Epstein
and Zin (1991) assumes that an aggregate equity return measures the return
on an aggregate wealth portfolio. In this case the continuation value relative
to a risk-adjusted counterpart that appears in formula (11) is revealed by the
return on the wealth portfolio for alternative choices of the preference param-
eters. Thus there is no need for an econometrician to compute continuation
values provided that data are available on the wealth portfolio return. Epstein
and Zin (1991) applied GMM methods to estimate preference parameters and
test model restrictions by altering appropriately the approach in Hansen and
26
Similarly, many models with heterogenous consumers/investors and incomplete mar-
kets imply pricing relation (1) for marginal agents defined as those who participate in the
market over the relevant investment period. Such models require either microeconomic
data and/or equilibria solutions computed using numerical methods.
6490_Book.indb 418
11/4/14 2:30 PM
Uncertainty Outside and Inside Economic Models 419
Singleton (1982). Given that the one-period SDF can be constructed from con-
sumption and return data, the full investor information set does not have to
be used in the econometric implementation.
27
Campbell (1993) and Campbell
and Vuolteenaho (2004) explored a related approach using a log-linear approxi-
mation, but this research allowed for market segmentation. Full participation
in financial markets is not required because the econometric specification that
is used to study the risk-return relation avoids having to use aggregate con-
sumption. Like Epstein and Zin (1991), this approach features the return on the
wealth portfolio as measured by an aggregate equity return, but now prospective
beliefs about that return also contribute to the (approximate) SDF.
4.3 a Continuing Role for GMM-based Testing
Even when fully specified stochastic equilibria are formulated and used as the
basis for estimation, there remains the important task of assessing the perfor-
mance of the pricing implications remains. SDFs constructed from fully speci-
fied and estimated stochastic equilibrium models can be constructed ex post and
used in testing the pricing implications for a variety of security returns. These
tests can be implemented formally using direct extensions of the methods that
I described in section 3. Thus the SDF specification remains an interesting way
to explore empirical implications, and GMM-style statistical tests of pricing re-
strictions remain an attractive and viable way to analyze models.
In the remainder of this essay I will speculate on the merits of one produc-
tive approach to addressing empirical challenges based in part on promising
recent research.
5 MiSSPeCiFied belieFS
So far I have focused primarily on uncertainty outside the model by exploring
econometric challenges, while letting risk averse agents inside the model have
rational expectations. Recall that rational expectations uses the model to con-
struct beliefs about the future.
28
I now consider the consequences of altering
27
In contrast to recursive utility models with ρ ≠ γ, often GMM-type methods can be
applied to habit persistence models of the type analyzed by Sundaresan (1989), Constan-
tinides (1990) and Heaton (1995) without having to specify the full set of information
available to investors.
28
A subtle distinction exists between two efforts to implement
rational expectations
in econometric models. When the rational expectations hypothis is imposed in a fully
specified stochastic equilibrium model, this imposion is part of an internally consistent
6490_Book.indb 419
11/4/14 2:30 PM