Thread by @AlistairMacaul7

Expectations of inflation/unemployment/etc feature heavily in policy (esp. this year).

But expectations aren't some universal variable that's the same for everyone – people disagree about macro variables all the time!

Should you care? My JMP shows you should!
#Econjobmarket

Specifically, I ask: when, and how, does heterogeneity in expectations affect the way shocks transmit to aggregate outcomes?

I look at that in a very general macro framework, that nests all your favourite GE models – and then I study an application to households and inflation.

But first, let's fix some ideas: to form an expectation, you need 2 things.

1) You start with some *information* on the values of some variables.

2) You pass that through a *model* to map from info to the variable you want to forecast.

Most existing models with heterogeneous expectations put the heterogeneity in *either* 1 or 2.

The big, general idea of my JMP is to show that the heterogeneity becomes much more important in shock transmission when you have heterogeneity in *both*.

And I can show you why with a simple example.

Suppose a government announces a large package of unfunded tax cuts. What effect does that have on currency markets?

(I'm a macroeconomist in 🇬🇧, can you blame me for picking this eg?)

Suppose for our e.g. that investors are split into two equal-size groups.

Group A believe that tax cuts strongly stimulate growth. They may or may not be right about that – that's just their subjective model of how those variables relate to each other.

If you tell Group A you're cutting taxes, they interpret that info as implying higher growth in the future. In our simple eg, they respond to that by increasing demand for £s – they want to invest in the soon-to-be growing economy!

But Group B have a different subjective model – they believe unfunded tax cuts have some damaging macro consequences (shocking, I know).

So if they see the announcement of a mini-budget, they respond differently – they panic, and reduce their demand for £s.

So heterogeneous subjective models ➡️ heterogeneous expectations (for growth in this case).

BUT: if both groups observe the announcement (same info), that heterogeneity is *irrelevant* for the aggregate demand for £s.

Why?

Group A demand ⬆️, group B demand ⬇️.

Aggregate change: average across the two.

You could exactly predict that change with just the *average* subjective model across groups.

Similarly, if everyone interprets tax information in the same way, but only some observe the mini-budget announcement:

Growth expectations: heterogeneous

Aggregate effect: summarized with just average info precision and average subjective model.

But if both info and models differ across groups? Suddenly averages aren't good enough.

Suppose A see the announcement, but B aren't watching UK news. Then A react, B don't. Aggregate £ demand⬆️

But if you reverse the info, so B are the well-informed ones? Aggregate £ demand⬇️

Those cases both have the same average info, and average subj. model. But different aggregate outcomes, because:

Heterogeneous models➡️heterogeneous responses to information. So systematic patterns in how info is allocated among those models affect aggregate shock transmission!

That example is very stark. Hopefully it may even seem a bit obvious!

The real work in the paper is to characterize that same effect in a very general class of macro models, with general distributions of info precision and general subjective models.

In that general class, I show you can always decompose the effects of an arbitrary shock on an aggregate variable into some standard terms, and a *cross-sectional covariance* between aspects of information and subjective models.

That covariance summarizes what I call the 'narrative heterogeneity channel': systematic relationships between heterogeneous info, and the heterogeneous models used to interpret it, can have big effects on the transmission of aggregate shocks.

So that's the general contribution. In the paper I discuss why you should expect to see that covariance != 0 under a range of popular theories of expectation formation.

I also study a topical application to household beliefs about inflation.

Briefly, I use unique features of a @BoE_Research survey to get individual-level measures of:

1) Does a household use any direct information about inflation when forming expectations?

2) How do they believe inflation is related to the real economy?

Unsurprisingly, there's a lot of heterogeneity in both! And the joint distribution of (info,model) changes over time.

I calibrate a model to that data, do my decomposition, and find that the transmission through narrative heterogeneity is large, and time-varying.

The key reason: households with lots of info about inflation are mostly those who believe inflation➡️ very damaging to the real economy.

So the people who see rising inflation first are the ones who hate it the most. That lowers the consumption response to inflation by a lot!

In the paper: much more on implications in this context, including

-Information selection effects
-Central bank communication
-Implications for survey RCTs (a la @YGorodnichenko)
-Should we worry about high inflation getting 'baked in' to expectations?

But this is just the first application.

The narrative heterogeneity channel I introduce here is much more general – this is just the start of a research agenda exploring its effects in other contexts, with other agents.

Thanks for reading! If you want to know more about this or my other research, check out my website: sites.google.com/site/alistairm…

And read the paper: amacaulayecon.github.io/WebFiles/Macau…

Also, go look up the incredible group of economists also on the market from @OxfordEconDept this year!

@AltmannM @bzdiop @shihangh @luke_milsom @HannahZillessen @HughLi85363603 @MBertazzini @VictorPouliquen Xiyu Jiao