Lars Peter Hansen Prize Lecture: Uncertainty Outside and Inside Economic Models

434 

The Nobel Prizes

models differently in good versus bad times. For instance, persistence in eco-

nomic growth is welcome in good times but not in bad times. Given ambiguity 

about how to weight models and aversion to that ambiguity, investors’ worst-

case models shift over time leading to changes in ambiguity price components.

Introducing uncertainty about models even with a unique prior will amplify 

risk prices, although for local risk prices this impact is sometimes small (see 

Hansen and Sargent (2010) for a discussion). Introducing ambiguity aversion 

or a concern about model misspecification will lead to a different perspective 

on both the source and magnitude of the market compensations for exposure to 

uncertainty. Moreover, by entertaining multiple models and priors over those 

models there is additional scope for variation in the market compensations as 

investors may fear different models depending on the state of the economy.

41

A framework for potential model misspecification also gives a structured 

way to capture “over-confidence.” Consider an environment with multiple 

agents. Some express full commitment to a benchmark model. Others realize 

the model is flawed and explore the consequences of model misspecification. 

If indeed the benchmark model is misspecified, then agents of the first type are 

over-confident in the model specification. Such an approach offers a novel way 

to capture this form of heterogeneity in preferences.

What is missing in my discussion of model misspecification is a prescription 

for constructing benchmark models and/or benchmark priors. Benchmarks are 

important for two reasons in this analysis. They are used as a reference point for 

robustness and as a reference point for computing ambiguity prices. I like the 

transparency of simpler models especially when they have basis in empirical 

work, and I view the ambition to construct the perfect model to be unattainable.

7 CoNCluSioN

I take this opportunity to make four concluding observations.

1.  The first part of my essay explored formal econometric methods that 

are applicable to a researcher outside the model when actors inside the 

model possess rational expectations. I showed how to connect GMM es-

timation methods with SDF formulations of stochastic discount factors 

41 

See Collin-Dufresne et al. (2013) for a Bayesian formulation with parameter learning 

that generates interesting variation in risk prices. Given that recursive utility and a pref-

erence for robustness to model misspecification have similar and sometimes identical 

implications for asset pricing in other settings, it would be of interest to see if this simi-

larity carries over to the parameter learning environments considered by these authors.

6490_Book.indb   434

11/4/14   2:30 PM

---

Uncertainty Outside and Inside Economic Models 435

to estimate and assess asset pricing models with connections to the mac-

roeconomy. I also described how to use SDF formulations to assess the 

empirical implications of asset pricing models more generally. I then 

shifted to a discussion of investor behavior inside the model, perhaps 

even motivated by my own experiences as an applied econometrician. 

More generally these investors may behave as if they have distorted be-

liefs. I suggested statistical challenges and concerns about model mis-

specification as a rationale for these distorted beliefs.

2.  I have identified ways that a researcher might alter beliefs for the actors 

within a model, but I make no claim that this is the only interesting way 

to structure such distortions. Providing structure, however, is a prereq-

uisite to formal assessment of the resulting models. I have also suggested 

statistical measures that extend the rational expectations appeal to the 

Law of Large Numbers for guiding the types of belief distortions that are 

reasonable to consider. This same statistical assessment should be a valu-

able input into other dynamic models within which economic agents 

have heterogeneous beliefs.

3.  How best to design econometric analysis in which econometricians and 

agents formally acknowledge this misspecificaton is surely a fertile av-

enue for future research. Moreover, there remains the challenge of how 

best to incorporate ambiguity aversion or concerns about model mis-

specification into a Marschak (1953), Hurwicz (1962) and Lucas (1972) 

style study of counterfactuals and policy interventions.

4.  Uncertainty, generally conceived, is not often embraced in public dis-

cussions of economic policy. When uncertainty includes incomplete 

knowledge of dynamic responses, we might well be led away from argu-

ments that “complicated problems require complicated solutions.” When 

complexity, even formulated probabilistically, is not fully understood by 

policy makers, perhaps it is the simpler policies that are more prudent. 

This could well apply to the design of monetary policy, environmental 

policy and financial market oversight. Enriching our toolkit to address 

formally such challenges will improve the guidance that economists give 

when applying models to policy analysis.

ReFeReNCeS

Ai, Chunrong and Xiaohong Chen. 2003. Efficient Estimation of Models with Condi-

tional Moment Restrictions Containing Unknown Functions. Econometrica  71 

(6):1795–1843.

6490_Book.indb   435

11/4/14   2:30 PM

---