Wednesday, January 28, 2009

The thing about economics ...

Humans are creatures of hope. It occurs to me that we place too much blind faith in things sometimes. Take economics. The thing about macro models is that they depend on input that has been smoothed and homogenized far below the noise levels of real life. Sort of like smoothing loan collateral pool risk into simplified spreadsheet factors. Economics too often depends on the systems finding balance presuming computational isolation from discontinuitues, externalities and outliers. The law of unintended consequences always applies when we simplify. Murphy's leverage is amplified by the alchemy of theories reaching far beyond their stable linearity. When you hear someone say "it's not perfect but it's better than nothing" that's the sound of a match being lit in a room full of dynamite. Sometimes we forget that we are the monkeys with keyboards trying to write Shakespeare.

So here's one for you to ponder. The same colleges that train some of the best economists also train some of "the best" structured finance designers. It's the same math. I'm just sayin'.

- D


  1. Amen, Dennis. As Eric Beinhocker (The Origin of Wealth) points out, the monkeys can try to write Shakespeare; but that does not mean they have any chance of success.

  2. Yes, but there's an oversight in representing the oversight as 'fat tails' of risk analysis like Nassim Talib does with his "black swan" model. It's correct except for the divergences from the normal distribution not being random. Divergences that matter to an otherwise reasonable model are *developmental*. That when you get caught making ever higher bets for the performance of realities that are falling ever further behind is the problem. It *looks* like statistical error from the view of the information model and the betting charts, but from the physical environment the divergence looks like ¸¸¸.•´ and gives you a lot of lead time to respond.

  3. Take noisy small sample set data. Winsorize the heck out of it to make it look Gaussian. Then feed it into a model that still delivers a 0.1 R^2. Anyone want to bet you life on that?

    And anyone can make an anomaly appear. All you have to do is not turn off your Monte Carlo program and wait. It'll look like something's growing but it's just the chip repeating itself because it has a finite count limit. Rocks aren't very smart no matter how much electricity you force feed them.

    Greetings Dr. Falken.
    Would you like to play a game?

    - Dennis