🚀 go-pugleaf

pcollac@pyrnova.mis.pyramid.com (Paul Collacchi) writes:

>In article <thompson.714869497@daphne.socsci.umn.edu>,
>thompson@atlas.socsci.umn.edu (T. Scott Thompson) writes:

>|> I am always amazed at how people claim to understand business cycles
>|> so well when most economists are of the opinion that they are among
>|> the most poorly understood of economic phenomena.

>I think its important to clarify and expand on this a little.  I think,
>by "understand" it is better to say "predict", since prediction would
>be the ultimate demonstration of understanding.  Secondly, I think most 

See below for some remarks on this.

[stuff deleted]

>|> I am _not_ suggesting that my alternative theories are right and
>|> others are wrong.  Instead, I am suggesting that the evidence about
>|> "causality" for business cycles is pretty weak.

>I'm a little surprised by this.  Aren't there mathematical models which
>are able, at least, to simulate GDP cycles?  These GDP cycles may not
>have amplitudes or frequencies which match anything real, but the models
>themselves may suggest something about the nature of variables at work 
>needed to produce curves that 'look like' GDP graphs.

The point is that there are _many_ mathematical models that are able
to do a pretty good job of simulating business cycle phenomena.  They
involve quite different causal mechanisms.  So simulation success is
not enough to resolve the disputes among the competing theories.

As for prediction...

I would agree that predictive success is in principle a pretty good
yardstick for evaluating models, but only if there is enough variation
in the model inputs so that we are able to test predictive success
under a wide range of scenarios.  The problem is that we can not
experiment much with the macro economy, so the models are rarely
tested (through prediction) in situations that are able to sort out
the differences among them.

For example, two of the mechanisms (among many) that might explain the
boom of the 1980s are (1) lower tax rates and (2) deficit spending.
In order to learn the relative contribution of these two phenomena we
would have to run some experiments in which we vary tax rates and
fiscal policies _independently_ holding all other things equal.
Needless to say, this cannot be done.

Some economists sometimes claim that "natural experiments" have
occured in the real economy.  Usually when you look closely at these
claims you find that they are based on a whole range of implausible
assumptions.

Bottom line: Predictive success is a good measure of the quality of a
model.  Measuring predictive success for business cycle models under a
reasonable varieety of scenarios is virtually impossible.

For what it's worth: My impression is that most of the formal
mathematical models that have been developed have a very poor track
record predicting movements of the business cycle more than a few
months ahead.  The most successful forecasting models do not attempt
to model causal mechanisms directly, and therefore are rather
uninformative about what the "fundamental forces" are driving the
business cycle.
--
T. Scott Thompson              email:  thompson@atlas.socsci.umn.edu
Department of Economics        phone:  (612) 625-0119
University of Minnesota        fax:    (612) 624-0209