[PREV - WHICH_WAY]    [TOP]

TOY_WORLDS

                                             May  9, 2008

                                             BLACK_SWAN

  "The only comment I found unacceptable was,
  'You are right; we need you to remind us
  of the weakness of these methods, but
  you cannot throw out the baby with the
  bath water," meaning that I need to accept
  their reductive Gaussian distribution while also
  accepting that large deviations could occur --
  they didn't realize the incompatibility of the
  two approaches."  -- p. 281

  Taleb sounds like quite the fanatic here.

  In the physical sciences, defective
  mathematical models are often used
  with the full knowledge that they're
  defective, in hopes that they might
  provide some insight in any case, and
  perhaps might even be patched somehow
  to achieve accuracy.

  Similarly, someone trying to deal with a large
  amount of short term economic data might very well
  prefer to attempt to use some grossly simplified
  models in hopes that you can get some qualitative
  sense of what's going on -- until something odd
  happens that invalidates the model completely, as
  indeed, is always possible with these models.

  I would submit -- as I gather many
  people have already -- that in the
  absence of anything better, it is by no      There is the difficulty
  means clear that one is better off           that a number generated
  using nothing (intuition alone, rather       from a mathematical
  than intuition guided?).                     model can have a problem with
                                               ambiguous precision.

                                               You might cite it to four
                                               significant figures,
                                               because it is a precise
                                               number, *given the
                                               assumptions*.

                                               But since you're not
                                               supposed to take the
                                               assumptions as
                                               givens, really,
                                               those four digits
                                               are probably three
                                               (at least) too many.

                                                   Among the hard sciences,
                                                   using numbers to express
                                                   a qualitative idea is
                                                   strongly frowned upon
                                                   ("I feel 80% sure that
                                                   I am right.")

                                                   By that standard,
                                                   publishing the
                                                   results of an
                                                   experimental model
                                                   that's *known* to be
                                                   defective does not
                                                   look so benign.


--------
[NEXT - KNOWN_QUANTITIES]