Physics Envy in Development (even worse than in Finance!)

  /  

Andrew Lo and Mark Mueller at MIT have a paper called "WARNING: Physics Envy May Be Hazardous to Your Wealth," also available as a video.  The takeaway, which is equally relevant to Development as to Finance (the actual topic of the talk),  is that inability to recognize radical UNCERTAINTY is what leads to excessive confidence in mathematical models of reality, and then on to bad policy and prediction.

Imagine how much harder physics would be if electrons had feelings! (Richard Feynman)

The key concept of the paper is to define a continuum of uncertainty from the less radical to the more radical. You get into trouble when you think there is a higher level of certainty than there really is.

  1. Complete Certainty
  2. Risk without Uncertainty (randomness when you know the exact probability distribution)
  3. Fully Reducible Uncertainty (known set of outcomes, known model, and lots of data, fits assumptions for classical statistical techniques, so you can get arbitrarily close to Type 2).
  4. Partially Reducible Uncertainty (“model uncertainty”: “we are in a casino that may or may not be honest, and the rules tend to change from time to time without notice.”)
  5. Irreducible Uncertainty:  Complete Ignorance (consult a priest or astrologer)

Physics Envy in Development leads you to think you are in Type 2 or Type 3, when you are really in Type 4. This feeds the futile search for the Grand Unifying Theory of Development.

Type 4 "model uncertainty" seems even more likely in development than in finance, theory is a lot better developed and produces more precise hypotheses in the latter than in the former (but even I am not so skeptical to think that we are in Type 5 in development).

What to do about large uncertainty in development? Obviously not a question that can be answered in one sentence.  Maybe we can start by discussing social systems that allow decentralized agents to solve their own problems that feature less uncertainty, and doesn't require any centralized agent to know the uncertain whole model of the whole system.