Day 4 #RWRI 14
1/n
Today we continued talking about fat tails and convexity. @financequant's talk had to be postpone because power went out in his location.
Personal highlights
- We have said before that we can't forecast fat tail variables. But why?
1/n
Today we continued talking about fat tails and convexity. @financequant's talk had to be postpone because power went out in his location.
Personal highlights
- We have said before that we can't forecast fat tail variables. But why?
2/n
Ans: Because fat tails are determine by one or a few points. The fatter the tail, the bigger the effect of one observation.
- This argument also explains why correlation has no meaning under fat tail variables. The relationship between two fat tails is given by a point
Ans: Because fat tails are determine by one or a few points. The fatter the tail, the bigger the effect of one observation.
- This argument also explains why correlation has no meaning under fat tail variables. The relationship between two fat tails is given by a point
3/n
that you probably don't even have. Hence correlation is meaningless, or as Nassim said the first day, corr is not corr. Here's a paper by Nassim related to this.
academia.edu/39797871/Foole…
- Only produce forecasts where the LLN applies.
that you probably don't even have. Hence correlation is meaningless, or as Nassim said the first day, corr is not corr. Here's a paper by Nassim related to this.
academia.edu/39797871/Foole…
- Only produce forecasts where the LLN applies.
4/n
- Let's return to convexity to answer the following:
Q. How do you make money with it?
A. Find places where people don't see the convexity.
Q. How to do business?
A. Look at how much you can lose and go backwards.
Be risk aware.
Convexity is the ultimate advantage
- Let's return to convexity to answer the following:
Q. How do you make money with it?
A. Find places where people don't see the convexity.
Q. How to do business?
A. Look at how much you can lose and go backwards.
Be risk aware.
Convexity is the ultimate advantage
5/n
- If you have a convex payoff, what matters are the second order effects, not the avg.
- Anything that is concave is eventually going to blow up.
- In Mediocristan one event won't make you reach. In Extremistan, one can. Related to day 2
- If you have a convex payoff, what matters are the second order effects, not the avg.
- Anything that is concave is eventually going to blow up.
- In Mediocristan one event won't make you reach. In Extremistan, one can. Related to day 2
https://twitter.com/mmencherogarcia/status/1293234043014848514?s=20
6/n
- The masquerade problem:
A power law can masquerade as a Gaussian. A Gaussian can't masquerade as a power law.
- Given n observations, can you rule out Gaussian?
No, unless you're God or you know the problem's constraints (eg caloric consump. Can't have more than x cal)
- The masquerade problem:
A power law can masquerade as a Gaussian. A Gaussian can't masquerade as a power law.
- Given n observations, can you rule out Gaussian?
No, unless you're God or you know the problem's constraints (eg caloric consump. Can't have more than x cal)
7/n
This is the black swan problem. You see a million white swans and can't rule out black. You see 1 black swan, then you can conclude that not all swans are white.
- Thin tail + fat tail => fat tail
- Latest paper by Nassim on fat tails:
forecasters.org/blog/2020/06/1…
This is the black swan problem. You see a million white swans and can't rule out black. You see 1 black swan, then you can conclude that not all swans are white.
- Thin tail + fat tail => fat tail
- Latest paper by Nassim on fat tails:
forecasters.org/blog/2020/06/1…
8/n
- Wittgenstein's ruler: It's vastly more likely that you have the wrong model than that a 10 sigma event has happened.
- Diego Zviovich presented some interesting models for epidemiology using cellular automata. Power went out on his location, but he continued by phone.
- Wittgenstein's ruler: It's vastly more likely that you have the wrong model than that a 10 sigma event has happened.
- Diego Zviovich presented some interesting models for epidemiology using cellular automata. Power went out on his location, but he continued by phone.
9/n #RWRI 14
So far 3 instructors have gotten disconnected at some point because power has gone out in their locations. And Nassim's laptop broke.
Thankfully #RWRI is anti-fragile :)
So far 3 instructors have gotten disconnected at some point because power has gone out in their locations. And Nassim's laptop broke.
Thankfully #RWRI is anti-fragile :)
• • •
Missing some Tweet in this thread? You can try to
force a refresh
