8 March 2008

Blowed up real good

March 07, 2008

"Stability is destabilizing" - Hyman Minsky

Before I became a wheeler-dealer online stock trader (eight years paycheck free!), I made my living building and running large computer models of oceans and atmospheres. The bigger and more complex the model, the greater the chance of "something going wrong" (i.e. the model becoming numerically unstable). This means some or many of the model values would go to infinity, whereupon the programs would halt.

It was a numerical, not a physics, problem. Though journal editors didn't like me describing how some particular model had "blowed up real good", they had started generating values (of say velocity) which went exponentially larger in an explosive way.

Typically, just before the models would go berserk, they would start exhibiting unusual behaviour. This meant they would develop swings in value, both in time and in "neighbouring" grid points. The swings would amplify and then something else would go unstable. Things that shouldn't correlate (move together) often started to quite strongly.

All manner of models are susceptible, and making them larger often didn't help. Global fluid models can go unstable just as easily as closed ones.

So, I would then look over the models, and then try to enhance, damp, smooth, decorrelate, or whatever, to try and get the models to work. It wasn't easy, and certain aspects of fluid modelling are considered insoluble. You just have to try and minimize the damage and get the things working.

When I left the field after cashing in on (or more correctly out of) the Tech Boom of the late nineties, I thought I was done with numerical instability. Which brings us to the main point; timeseries of financial instruments are starting to look a lot like those fluid models about to go unstable.

Bonds drop a point in one day, recover the next, silver drops 30 cents as funds exit en masse, and so on. The financial model analogy is not so far fetched. There are hundreds, if not thousands, of computers doing dynamic risk hedging every day of the week, all interacting and buying and selling protection of one sort or another.

There are hedge funds trading hundreds of billions of dollars in a heartbeat waiting to jump on the next trend. Governments of all size are trying to "do something" to manipulate their currencies and respective economies. It's a potent combination, and getting bigger every day. And previously "uncorrelated" financial instruments of many kinds suddenly are. Quite strongly.

One of the recent advances made in fluid numerical modelling was stability analysis. You'd look at the equations and the way they were implemented and look for unstable "modes". This was often done after a particular mode had made it's presence known in a more painful way. And, as the models got more complex, so did the stability analysis to the point where it really wasn't of much help.

The models I ran could therefore go unstable purely by their own internal structure. Run along fine for a while, and then,"pffft", off they'd go. No external shock was needed to get the ball rolling. Equivalently, the 10 sigma event which may create financial havoc may lie within.
In ocean modelling, it is possible to smooth out bottom topography, limit changes in model variables to small increments, and force conditions back to "long term averages". The price paid for such levels of adjustment is that you are effectively introducing fictitious forces, and the models then often bear little resemblance to reality.

Financial markets, unfortunately, also look a lot like these kinds of models at times. The heavy hand of stability has been all too evident in the financial markets (e.g. volatility indices) the past few years, to the point where the "real" economy is probably totally skewed.

As an another example, in the book "Competion" by James Case, there is a reproduction of Senator Moynihan's argument against the balanced budget proposal of the 1980's. It shows 100 years of US GDP growth smoothed out after WW II with the introduction of Keynesian stimulus. This has been trumpeted as a good thing. I wonder. When I did similar things in the models, it oftentimes meant that the imbalances responsible for the natural variability built up over time. Then they became a bigger problem.

The ultimate cause of the problems in fluid models is limitations of computer size and speed . You have to represent the aggregate of all the fluid processes in a particular (often spatially large) grid box to a few numbers. The real world "sorts itself out" on smaller scales not resolved by the computer models.

The equivalent financial problem today is too much wealth and power in too few hands. The system accomodates the accumulation of great wealth, and legislation designed to prevent financial excesses has all but been eliminated.

There are now few "independent" economies or other anchors (like gold used to be) which could counterbalance a disruption in another area. And this hugely macro economy has been stabilized all too well to the point where the building imbalances may ensure that "something's gotta give".

And how, pray tell, does this help us make better investment decisions? Well first, you can probably count on a whole lot more volatility ahead. It isn't a time to be sticking your financial neck out to the point where a sudden wild swing can wipe you out. One strike and you're out in this league. Margin bad, liquidity good.

Dave Ramsden

Second, the financial "mode" which ultimately "goes exponential" may never be fully identified. There's no point trying to understand potential causation. "Safe" financial strategies may backfire as easily as "risky" ones. Stay flexible.

Third, big models are the sum of a lot of smaller, separate processes. After a period of huge frustration, we'd often break the models down into simpler, localized versions and try to understand and cure the problems there. Often, that was all we could do. Some bigger models never did work.

Led by cheap energy and technological advances, a long period of economic integration is now under stress. I see a potential process of dis-integration ahead. Lots of "paper promises" by large institutions are breaking down. There've been recent recent bank runs and halts on fund redemptions. The underlying economic model shows signs of breaking down (or breaking up).

Big institutions are under pressure to prevent a global meltdown. Personally, I see the unfolding crisis as beyond anyone's understanding or control. That's certainly what the timeseries data says to my eye. How bad the crisis will get, what form it will take, or how long to unfold are anybody's guesses, but I wouldn't count on government to supply your basic needs in ten years or so.

So, be prepared for a return of the 70's catch phrase "small is beautiful", only in this case, it might be more akin to "small is all we got". Who knows how local economies will need to get to reach sustainable equilibria.

And, if things get really bad, prepare yourself to participate in a very local economy, so have resources close to hand. Your financial institution may not answer the phone one day, so have some of your assets in real, physical form.

The adjustment to a new economic world will take a while. Problem solvers, like investors, are going to have to come to their senses one at a time. Big economies are ultimately built out of small, resilient ones, and a lot of competitive smaller players would really help the process.

Maybe by then there'll be a new macro paradigm to replace the discredited ones, with even some Minsky baked into official economic models. Ones that won't blow up real good.

No comments: