33  A Workflow for Dangerous Data

Distills the entire book into a checklist meant to be pinned above a desk:

Get the data
   -> Plot it (histogram + log-log survival)
   -> Ask which moments plausibly exist
   -> Check dependence, especially in the extremes
   -> Model the tail explicitly if it matters
   -> Stress the estimator (resample, perturb, vary k)
   -> Quantify uncertainty honestly
   -> Only then make the decision

33.1 Chapter 32: What We Can Actually Know

An epistemic chapter: sometimes two very different tail models (say, a lognormal and a Pareto) fit the available data almost equally well, and no amount of clever estimation resolves that ambiguity without more data — occasionally more than will ever exist. The honest answer, the chapter argues, is sometimes “we can’t tell yet,” and pretending otherwise is a bigger risk than the uncertainty itself.

33.2 Chapter 33: Go Forth and Look at the Tail

Closes not with a library tour but with a single instruction meant to outlast the book: the next time someone hands you an average, ask what happened in the tail before you believe it. Pointers to Mandelbrot’s work on scaling in markets, Taleb’s Statistical Consequences of Fat Tails, and the extreme-value-theory literature are offered for readers who want to go from aware to rigorous.