3 Comments

Hi Daniel: Engin Yilmaz forwarded this post to me; I find it very clear and coherent. (And the whole blog is very interesting.)

I wrote what ended up being a kind of lengthy reply, and I though I'd share it with you here. Thanks and cheers.

Very useful and short understanding of macro-model frameworks, how to impart them clearly and coherently.

Basically, what's your economic model's "story" of how the economy works, and how does that story look, precisely, in sources and uses T accounts? Is it accounting-coherent?

How does your story divide the economy into "sectors" (or often, "functional categories" of actors/institutions, which are often orthogonal to any national-accounts sectors). And what sectors are absent?

What objects (assets) are included (and not) within the T-account cells? Do they model transfers of M2, bonds, real (long-lived) goods, etc.?

What are the relationships between the sectors, and the measures therein, as played out in the T accounts? When actors in sector X do Y, what changes are caused, in what other sectors?

The T accounts give precise form to the narrative description of the story, and let others see in condensed form what's included, what's missing, and how the included items relate.

Focusing a bit on those "functional categories" used in many models: As an inveterate "Show-Me-The-Numbers" guy from Missouri, I mostly find these "functional" categories less than useful, because I can't go look at the time-series for them in national accounts.

Researchers/model-builders are forced to effectively build their own national-accounts "sectors," and assemble time series for them from disparate sources. These series:

1. Are often/mostly not published with the model, at least in tractable form.

2. Are often missing precise, accounting-coherent descriptions of their derivations and mutual identities.

3. Are generally somewhat idiosyncratic (and not-infrequently, erroneous). It takes some serious work to unpack that down to the level of time series.

More generally: they don't provide a good basis for the larger macro-modeling conversation, based on published, carefully defined and documented series that in toto are accounting-identity coherent. With very few exceptions, those sets of mutually-coherent time series are and almost must be issued by national accountants. 

PSZ tables are one good counter-example to these objections. There are certainly others. But understanding their relationships to national accounts tables requires some seriously lengthy and rather excruciating yeoman's work. I'm here to tell ya. ;-)

So I would suggest that modelers should think long and hard before building models, T accounts, and their associated stories that are not tightly linked to generally-available and broadly "legible" national-accounts sectoral series. 

This is basically suggesting a fundamentally empirical limit on modelers' imaginations. In general, try to build models in which the time series do not have to be heavily massaged, and a whole novel "sector" assembled — with all the opaqueness and at least potential error that inevitably result.

FWIW...

Expand full comment
author

Hi Steve, and thanks for reading!

To try to put it concisely, I think I am trying to pull accounting frameworks out as a meta-category for talking about other macro models. I don't have a particular story about the relationships among macroeconomic variables. Instead I think that different stories apply at different times, so it is helpful to keep them talking to one another. A lot of macro is very disconnected from finance, and I also find it is helpful to be able to identify that, and sometimes to correct it.

I would also say that there is a two-way relationship between theory and measurement: yes we should try to theorize about what we can measure, as you say, but also what gets measured is driven by theory. We spend a lot of time on NIPA accounts in part *because of* decades of theoretical investment in them. So we do have an obligation, I think, to not just accept what is given. Maybe a better theory would lead to better data. The OFR is a recent example of this, on a small scale perhaps.

Expand full comment

Hey Daniel, thanks, agree with all. Thinking this through out loud, hope you don't mind...:

The accounting, it seems, is a *constraint* on models; it can disprove models, on their own terms.

The model's accounting must be coherent — all identities must hold when the actual measured, observed numbers are examined. And in a really full macro model like G&L's Chap. 11 Growth Model prototype or Zezza and Nikoforos' models, it must be ~complete. Big, empirically important measures must not be ignored. (This is me pounding my spoon on my usual high chair about accrued holding gains, a quite massive and mostly ignored measure, so the need for Haig-Simons income.)

But very much yes: The story itself determines what the accounting identities will be. The web of identities is just a definition of terms — a precise (and hopefully coherent and complete) embodiment of the story.

So saying "it's an accounting identity!" (grrrr) proves nothing; the identities and the story they arise from might be problematic. But we can say: if a model is self-contradictory, accounting-incoherent in its own identity terms, there's a problem with the story.

Or maybe more simply: the accounting construct is itself an economic model, embodying a story about how the economy works. And yeah there's no "true story" out there. There are only more and less coherent (and complete) ones.

Cheers,

Steve

Expand full comment