Jan

7

Height, weight, are distributed normally. Once an infinite variable like wealth is introduced, the distribution is no longer normal and can't be regressed. Lots of implications including convexity here.

Past extremes are not good predictors of future extremes. (The Lucretius Fallacy) Simple proof is that the last biggest was bigger than the one before.

Nils Poertner writes:

markets sometimes go from one extreme to another, they tend to overshoot like a novice at the sailing boat - always over-steering the boat. great for us as trader/investor.

William Huggins comments:

They can be regressed but only after an appropriate transform (log, ln, etc). The key is transform in reverse before interpreting outcomes.

If the data is time series though, you'll find that exponential growth (organic) results in "exploding variance' that makes the coefficient estimates less reliable (larger standard errors). Feasible generalized least squares maybe more practical than OLS in such cases.

Zubin Al Genubi responds:

A transform is the standard work around but you can't transform an infinite variable into CLT compliance without losing so much important information to make it dangerous. Its a different distribution. That's my point.


Comments

Name

Email

Website

Speak your mind

Archives

Resources & Links

Search