Taleb has a good video on fragility and the assymetry of risk. The video is 68 minutes long and covers quite a few things involving the assymetry of risk, anti-fragility and robustness.
One point is well made about Jensen’s inequality and the nature of variance about the mean which significantly alters tail risk.
The way I think about these things is that a complex system has multiple components. Those components variances are not necessarily simplisticly additive. The potential for interaction to occur due to some form of coupling induces uncertainty in the systems outcomes. The more components in the systems the greater the uncertainty of the outcomes. Additional portfolio diversification comes at the expense of knowing what is in the portfolio. If you don’t understand the sources of value or return, don’t add more of them.
This sounds obvious, but most people don’t quite get it. The addition of multiple components can vastly increase the number of “hidden” paths the overall system may take, these hidden paths include extreme tail variance.
Many systems and risks are assymetric in that thier positive variance is either de-minimus or trivial. An airplane is a complex system made of multiple parts, those parts or components may each have a known failure rate, but the fact is that the overall failure rate of the large system is likely greater than the cumulative components failure rate, due to unforeseen systemic interactions. The asymmetry of the system variance is obvious. A normal functioning plane flys, a failed system crashes.
Interestingly a rule of complex systems is that they are almost never succesffully designed functional from scratch, but rather evolve from simple small systems that function.
The word evolve is logical here as one can think of a complex organism such as the human is made up of over 200 kind of specialized and differentiated/evolved cells which evolved over time into differentiated functions which allowed for a net symbiotic gain in individual reproduction.
From a risk perspective the rules fall out as the following.
- 1. Component risk is not additive at the system level.
- 2. Historical system risk and output behavior is inversely useful relative to the number of component connection/links in the system.
- 3. Asymmetrical risk in systems must be managed by adding capacity / tolerance for failure or minimizing / dampening the components interaction.
- 4. Be aware of the trade off between efficiency and robustness across components and across longer time horizons.
For robust design of a banking, biological, political, manufacturing etc. system This boils down, to keep it simple, keep it small, keep it buffered.