An interesting paper has been published showing that many extinction events in nature happen due to correlated exogenous phenomenon which looks like colored environmental noise. This is interesting as most biological models had assumed that environmental (outside the species) variables were random, ie. white noise. The evidence indicated extinctions were accelerated due to correlated exogenous (environmental) noise. See link here.
This is important to econometric and finance models for the following reasons. Most quant risk models which are used to justify extra leverage or risk allocations for banks and traders are limited to historical market or nearby associated econometric data. The inherent assumption being that all prior data reflects enough prior noise and events to accurately reflect a noise profile for an instrument or portfolio of related risk.
Think about it like this. If I have a simple one variable non-interacting system that ranges from 1 to 100 over a given amount of time like a spring, I can be pretty sure that with enough samplings say X, I have seen the 1-100 range of likely events and their distribution. If I have a 2 variable system with loosely interacting components I may need a Y times X samplings to see the extreme ranges of possibilities.
One can see that extending the context of the system drivers increases dramatically the amount of data required to understand the range of outcomes. What isn't taken into account is the role exogenous harmonics might induce in the model. Even weakly interacting system can sync up with dangerous effects. For example, two grand father clocks set a few feet apart will eventually sync up. Consider that for a moment. Two systems could be set up to be non-correlated and yet over time will become exactly correlated, just like that portfolio last year experiencing a 25th standard deviation event. The event wasn't 25th standard deviation, just the way it was measured in the first place.
An environmental system rapidly becomes complex enough to by impossible to model relative to the drivers and historical data available. Even a simple closed predator prey network has vast complexity.
Economic systems or markets are the same way, in that the external variables make the pursuit of a true model challenging. We are currently undergoing a rapid economic shift that is a result of a credit bubble which is impacting markets far removed from the perceived point source.
The economic environment is going through a period of rapid accelerated extinctions due to these events. Yet many still operate under the assumption we can model our way through risk control and portfolio construction going forward using the same tools that got us here with a few tweaks.
Finance models fail due to an inability to capture context both at the physical and at the agent behavioral level. I have yet to see a risk model incorporate the risk that everyone else starts using the same model inducing over weighting and run away prices in an asset class or instrument.
After all most modern finance risk models are mostly trend following band wagon style into whatever was hot over the last year, decade etc. with a little tweak for co-variance, CAPM, Mean variance approaches at portfolio construction etc. Bandwagon finance as such isn't a thinking persons approach to risk or portfolio construction. Most applied risk allocation models are clever but useless tools. Perfect in a vacuum, but down right dangerous in the real world if you don't look at the assets underneath or the environment driving their price.
I remember studying Artificial Intelligence for fun in college and the professor discussed context in regards to a chess playing computer. The computer might be able to analyze deeply the possibilities of the chess board, but wouldn't know that it was time to leave the game if the room caught on fire. The context was limited and always would be by definition of a closed model. The inability of most risk models to incorporate variable context, regime shift or environmental state shifts,"such as NINJA loans" coupled with our over-reliance on them is a dangerous proposition and sadly one we haven't learned yet.
My suggestion would be to drop any mean variance portfolio or bank risk model which assumes to show via historical data the risk relationship between economic instruments. I advocate a return to simple leverage metrics and the elimination of VaR.
Basic rule of thumb leverage is in-efficient but stable, a trade off many would be happy to make right now. If it can't be understood within a traditional banking context don't touch it. Ignorance admitted isn't a bad thing, false pride is.
