Session 1: Questions for Prof Taleb

[RonRichman]

Please use this thread to submit questions for Session 1:

https://www.meetup.com/global-technical-incerto-reading-club/events/275697847/

[FergM]

Chapter 5: Zipf Plot

Question

How do you draw the tail fit line in Zipf plots?

Context:

I want to build a Python function for survival plots.

• Is OLS regression a bad idea?
• What's a better idea?
• Is the answer later in the book? 😄

Examples:

Chapter 5 Zipf plot:

Python Zipf plot code:

• https://github.com/FergM/fattails/blob/main/notebooks/NB-25 - Survival Plot - Gamestop.ipynb

[FergM]

Mathematica Code

Could you share some of the Mathematica Code?

[FergM]

Chapter 8: How much data do I need?

Question

Can you confirm whether Equation 8.8 is correct?

Equations

Book version:

My derivation:

My workings:

• https://github.com/FergM/fattails/blob/main/docs/Notes-02 - Derivation - How much data do I need.pdf
• If correct, my derivation would delay entry into the fughedaboudit zone 🤔

[rsnb]

Shouldn't the quote at the beginning of chapter 3 be:

תְּהֽוֹם־אֶל־תְּה֣וֹם ק֖וֹרֵא

instead of:

תְּהֽוֹם־אֶל־תְּה֣וֹם כוֹרֵא

?

[dylouie]

How can we apply (further) fat-tailed distributions in clinical research (clinical epidemiology) and biostatistics (i.e. survival analysis, clinical trials, and observational studies)? Or what limitations should doctors and epidemiologists acknowledge whenever they report results using their statistical models? I have been reading your posts on your comments on some Lancet journal articles, Lindy Effect, and using the entropy approach, and the "Tail Risk of Contagious Diseases", and I find them very insightful...

ADD: I am a medical doctor who has been studying a lot of epidemiology and biostatistics. Is it possible to apply the maximum entropy approach in statistical models for cohort studies and clinical trials? Thank you very much in advance!

[caiovinchi]

Chapter 25

First I would like to thanks Mr. Taleb and the organizers for this opportunity. Hopefully I got the line correct and my questions are not too amateur.

25.2.1 Analyzing the Constrains:

• can we first set epsilon at some level (say 0.99 or maybe other "ideal" level?) and then proceed to calculate VaR assuming normality (as per maximizing entropy condition) or perhaps should we calculate it after an EVT approach?
• it's there an impact on the risk of ruin or on the overall performance if the choice of K happens to be "too small" given the initial capital?

Proposition 25.1

• pluging the parameters in the mu formula for a simple B&H portfolio of two risky assets would often yield a negative value for epsilon around 0.5 or higher, does that mean the portfolio should avoid losses above this apparently "low" threshold? (so that the condition mu>0, sigma>K/n(e) are satisfied) and further dynamically adjusted via stop-loss/options?

Trying to execute the function in R under the global mean constraint would often yield a distribution with a huge spike, would that be the spot what which we set the Dirac function for the stop?

Thank you for your time!