In a pen-l debate, I wrote:
... A theory's "robustness" is very relevant. If minor variations of
a specification of a theory (i.e., the model used to summarize the
theory) lead to radically different results, that's quite an important
result, indicating that something's wrong with the theory. Maybe
something important. But if a robust model does not have any
theoretical content  (about causation and the like), that's a sign
that something's wrong, too.

By the way, unlike many economists, I distinguish between a "theory" (an idea about how 
causation works in the empirical world) and a "model" (a way of summarizing such an idea and 
communicating it to others, while testing its internal logical consistency). In econometrics, a 
theory's model may be specified in more than one way.<

this last part should be broken down into pieces:

an economic theory is an insight about how causation works in the
empirical world (as if there is any other world). For example, we
might assert that lowered expected real interest rates encourage
higher real fixed domestic investment at the macro level, based on
relatively simple economic logic.

an economic model is a way of summarizing the idea and communicating
it to others, while testing its internal logical consistency. There is
more than one way to "model" the theory of interest rates --> fixed
investment. For example, there's the Jorgenson neoclassical
(Cobb-Douglas production function) model with partial adjustment, but
there are other models.

By the way, I've noticed that among economists, the idea of a theory
is often conflated with that of a model. One told me that the main
"theory" of investment was the Jorgenson neoclassical one with partial
adjustment. It's like confusing words on the page (the model) with the
thought behind them (the theory)! Of course, many neoclassical
theories _are_ that shallow, mere mathematical shells without any
concern for underlying relations of causation. Such as: what in hell
is the causation story behind the Cobb-Douglas production function?

there is a third level, for econometrics. An economic model can be
stated as an econometric one -- a specification -- in order to test
it. Sometimes the economic model is much the same as the
specification: the Jorgenson model is a good case of this, I guess.
That would explain its popularity, as does its ideological content.

If I understand time-series analysis (ARIMA, ARMA, etc.) correctly, it
involves a specification without an economic theory (because there is
no story of causation behind it). It does not involve a true economic
model (as I define it) either. Julio, please tell me why this
conception of time-series analysis is incorrect.

If the specification doesn't pass the data test, that does not
automatically knock down the economic model, since other
specifications of the model can be found. It also doesn't knock down
the theory, since more than one model can be found for the same
theory.

However, it might provisionally tell us that the model and the theory
are weak and need to be better, until a better theory and a better
model come along. Unless we are willing to accept the assertion "who
the hell knows?", it takes a theory to trump a theory. A theory may
persist simply because there's not a better one available. However, it
should be taken with major dollops of salt.

If the specification _does_ pass the data test, that doesn't "prove"
the model or the theory. (Empirical proof is impossible.) However, if
there are no other theories available, it's a "reasonable" model and
theory. Again, it has to be compared to other theories and their
models.

--
Jim Devine / "Segui il tuo corso, e lascia dir le genti." (Go your own
way and let people talk.) --  Karl, paraphrasing Dante.

Reply via email to