Hi #EconTwitter!

A thread about my job market paper:

"Disagreement Aversion"

JMP link: ssrn.com/abstract=39641…

Although my other works document the political economy of climate change with empirical evidence (e.g. aeaweb.org/articles?id=10…), this paper is more theoretical.

1/20^{}

A thread about my job market paper:

"Disagreement Aversion"

JMP link: ssrn.com/abstract=39641…

Although my other works document the political economy of climate change with empirical evidence (e.g. aeaweb.org/articles?id=10…), this paper is more theoretical.

1/20

Experts disagree.

Take climate sensitivity, a parameter crucial to model climate change: climate modelers disagree on its probability distribution.

How should we aggregate these experts' distributions? How should we make decisions under expert disagreement?^{}

Take climate sensitivity, a parameter crucial to model climate change: climate modelers disagree on its probability distribution.

How should we aggregate these experts' distributions? How should we make decisions under expert disagreement?

The benchmark method is linear pooling: to evaluate any option, compute its expected utility using the average of experts' distributions.

Using linear pooling means being neutral to disagreement. It is a perfectly valid method, but we may want to be more cautious.^{}

Using linear pooling means being neutral to disagreement. It is a perfectly valid method, but we may want to be more cautious.

Models of ambiguity aversion (KMM, maxmin, robust control...) are commonly used to model a cautious aggregation of experts' beliefs: if an expert has more pessimistic views on (the expected utility of) an option, we listen to her more (to evaluate that option).
^{}

(Note that, as usual in the ambiguity aversion literature, we assume away strategic concern: all experts share the decision-maker's tastes; only their belief distributions differ.)
^{}

Ambiguity aversion only models aversion to disagreement regarding expected utilities.

Yet, there can be spurious agreement on expected utilities, masking underlying disagreement on the distribution of outcomes. Let us see this through an example.^{}

Yet, there can be spurious agreement on expected utilities, masking underlying disagreement on the distribution of outcomes. Let us see this through an example.

Take a CEO who hesitates between implementing a project or not. Both experts she consults are indifferent, but for different reasons. Expert 1 foresees no gain and no loss from undertaking the project; while expert 2 believes the chances of gain balance out the chances of loss.
^{}

Ambiguity aversion models aggregate experts' expected utilities (all equal to 0 here) and thus make the decision-maker indifferent to the project, just as the experts. They cannot capture a potential aversion to disagreement on the distribution of outcomes.
^{}

However, as experts disagree on the distribution of outcomes only in the case of the project, being disagreement averse would lead to prefer 𝘯𝘰𝘵 to implement the project.

Our paper makes three contributions.^{}

Our paper makes three contributions.

1. We introduce a novel notion of disagreement aversion, which pushes ambiguity aversion one step further.

Being disagreement-averse is: preferring the option with consensus on the distribution of outcomes compared to the option only featuring consensus on expected utilities.^{}

Being disagreement-averse is: preferring the option with consensus on the distribution of outcomes compared to the option only featuring consensus on expected utilities.

2. We derive a model that can account for disagreement aversion. We call it Distribution-Aggregating because, when evaluating an option alpha, it proceeds by aggregating (in a potentially cautious way) experts' Cumulative Distribution Functions, through an aggregator function I.
^{}

How does this compare to common ambiguity aversion models? Well, these models are all Utility-Aggregating. They also employ a (potentially cautious) aggregator function I, but at a different stage: they aggregate expected-utilities rather than CDFs.
^{}

We show that our DA model is more ambiguity averse than a corresponding UA model (for I concave). This is because only DA models can exhibit disagreement aversion.

We also provide an axiomatization of DA. The key is to relax UA's monotonicity axiom into monotonicity w.r.t. FOSD.^{}

We also provide an axiomatization of DA. The key is to relax UA's monotonicity axiom into monotonicity w.r.t. FOSD.

3. We draw implications of disagreement aversion in concrete applications: precautionary savings, and climate change.

In particular, we prove the general result for DA models, that increasing disagreement aversion leads to more cautious choices (more savings, more abatement).^{}

In particular, we prove the general result for DA models, that increasing disagreement aversion leads to more cautious choices (more savings, more abatement).

UA models don't share this intuitive property. To get a similar result that more ambiguity aversion leads to more cautious choices, UA needs an extra assumption: that experts' distributions are ordered in terms of First Order Stochastic Dominance (FOSD, i.e. optimism).
^{}

This assumption is very restrictive and generally does not hold in practice. Yet, this is the assumption that applied papers had to make when studying expert disagreement, e.g. @CGollier (ReStud, 2011); @AntonyMillner, Dietz & Heal (2013).
^{}

Not anymore. Our Distribution-Aggregating model together with the above result provides a simple, ready-to-use alternative to ambiguity aversion models.

We hope to see many applications, where expert disagreement matters: climate change, portfolio choice, savings decisions...^{}

We hope to see many applications, where expert disagreement matters: climate change, portfolio choice, savings decisions...

On the decision theory side, we contribute to the quest for models that go beyond expected utility (SEU) and can account for so-called paradoxes like Allais' or Ellsberg's.

The DA model is one of the few models that can account for both.^{}

The DA model is one of the few models that can account for both.

Thanks for reading!

For the sake of twitter, I departed from the paper in a number of shortcuts, approximations, and terminology changes.

If you want to know more about this paper and about my other research, you can check out my website here:

sites.google.com/view/adrien-fa…^{}

For the sake of twitter, I departed from the paper in a number of shortcuts, approximations, and terminology changes.

If you want to know more about this paper and about my other research, you can check out my website here:

sites.google.com/view/adrien-fa…

This paper is a joint work with fantastic colleagues from the @ETH: Antoine Bommier, Arnaud Goussebaïle, and @DanielHeyen.

In this collective endeavor, I was in charge of the Theory section of the paper: I am the one who proposed the setting and proved the propositions. 🤓

[end]^{}

In this collective endeavor, I was in charge of the Theory section of the paper: I am the one who proposed the setting and proved the propositions. 🤓

[end]

@threadreaderapp

unroll^{}

unroll

• • •

Missing some Tweet in this thread? You can try to
force a refresh

**
Keep Current with
Adrien Fabre
**

Stay in touch and get notified when new unrolls are available from this author!

**
This Thread may be Removed Anytime!**

Twitter may remove this content at anytime! Save it as PDF for later use!

- Follow @ThreadReaderApp to mention us!
- From a Twitter thread mention us with a keyword "unroll"

`@threadreaderapp unroll`

Practice here first or read more on our help page!

Support us! We are indie developers!

This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

**Become a Premium Member** ($3/month or $30/year) and get exclusive features!

Too expensive? **Make a small donation** by buying us coffee ($5) or help with server cost ($10)

Or Donate anonymously using crypto!

**Ethereum**

`0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E`

copy

**Bitcoin**

`3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi`

copy

Thank you for your support!

Check out other threads from @adrien_fabre!

Read all threads by @adrien_fabre