In case you're late to the party:
1/4 In the absence of a physical law forcing boundaries on a metric, it becomes fat-tailed, i.e. a single observation can be more important than everything that came before, combined.
1/4 In the absence of a physical law forcing boundaries on a metric, it becomes fat-tailed, i.e. a single observation can be more important than everything that came before, combined.
2/4 There is this parameter called alpha, which quantifies the thickness of the tail, i.e. how bad the situation is compared to one where you can happily just use normal approximations and non-parametrics.
david-salazar.github.io/2020/05/19/und…
david-salazar.github.io/2020/05/19/und…
3/4 Turns out that the alpha exponent is actually pretty well-behaved, that is, you don't need a ton of data to estimate it, and it gives you veeeeery important information as regards the actions you should be taking.
4/4 Preprint (by @nntaleb, @yaneerbaryam & @DrCirillo) uses everyone's favourite pandemic as a case example. It's standalone and worth re-re-reading, but if you want context, google "Taleb Ioannidis debate International Journal of Forecasting".
forecasters.org/wp-content/upl…
forecasters.org/wp-content/upl…
