Rumble! Re: Note to JB: (Time lags and casualties)

James Rogers jamesr at
Fri Sep 26 16:31:14 PDT 2003

On Fri, 2003-09-26 at 12:48, Joe Barrera wrote:
> James, I have to say, predicting recessions and booms
> 15 years in the future seems about as plausible as predicting
> rain-or-shine six months in the future. I mean, it's
> hard enough predicting chaotic systems when you can actually
> capture the state of the system and when the components
> of the system act deterministically...

Very different actually, and here is why:  The best long-term economic
predictors use people patterns (exquisitely modeled from hundreds of
different kinds of demographic data) as their primary data source from
which high-order algorithmic patterns are derived, which are then used
to derive macro-economic patterns.

Unlike a lot of other things you could measure, people patterns shift
very slowly, on the order of months and years.  Because of the extremely
low sample rate for this kind of data, ten or fifteen years does not
represent that many consecutive samples from the model function.  The
low sample rate and the fact that people have reliable behaviors in
aggregate means that you can predict derivative data points quite a ways
out with minimal divergence if you have enough historical data. 
Divergence does set in the older the model, and as I mentioned before,
the DotCom collapse happened a year earlier (plus or minus a quarter)
than predicted in the 1980s model that I saw.

The models I saw in the late 1980s did not say that the DotCom Boom was
going to occur specifically, just that there was going to be a people
pattern "perfect storm" that was going to create a frenzied bubble of
economic growth in the US economy.  The Internet just happened to be in
the right place at the right time, and if it wasn't that it would have
been something else.

Actually modeling the details of the economy and market patterns
directly is riding the edge of computational tractability.  You can only
do that with reasonable accuracy for short future windows because the
data resolution is so much higher.  Hence "tactical" models, which use
high-resolution data directly, versus "strategic" models which actually
infer macro-economic patterns from people patterns, which are coarser.

To think of it another way, both models have the exact same complexity
with the difference being that one sacrifices resolution of the
prediction for the range of the prediction.  Tactical models give
remarkably detailed predictions, but diverge from reality quickly (the
more detail they have, the faster they diverge). Strategic models make
only very coarse predictions, but diverge much slower.

It doesn't matter if you can model the system deterministically, only
that the system be deterministic at some level.

> P.S. Oh, and the Foundation Trilogy was *fiction* :-)

And pretty dry fiction at that.  I think I finished the first book
before losing interest, but that was a long time ago.

We can safely say that "psychohistory" as presented in that particular
book, while valid theoretically, passes the point of "computationally
intractable" some astronomical number of bus stops back for many


-James Rogers
 jamesr at

More information about the FoRK mailing list