Conversation

A bit of a definitional issue -- the interpretation is mostly wrong; the solution is the correct solution to *a* question but not to *the right question to be asking*

Dan Robinson

@danrobinson

Dec 2, 2020

Replying to

@SBF_FTX

and

@SBF_Alameda

What makes it the wrong question?

Replying to

It's trying to maximize EV[log(money)] instead of EV[money]. This is probably wrong in and of itself as an assumption to make. But if you *do* want to make that assumption, then e.g. you have to consider all of your assets that don't have anything to do with the pool.

Replying to

and

Do you disagree with the Kelly criterion in general? en.wikipedia.org/wiki/Kelly_cri

SBF

@SBF_FTX

Replying to

@danrobinson

So: 1) it's not literally *wrong* in that it doesn't make a math error 2) I disagree in general that Kelly is the right way for people to bet because I disagree with its assumptions 3) neither 1/2 are necessary for my argument: Kelly only works if you consider *all* your money

9:42 PM · Dec 2, 2020Twitter Web App

Replying to

and

There are other problematic assumptions snuck in here, e.g. a) the paper assumes that you can only do one or the other forever and can't ever take profit if you don't use an AMM b) the paper assumes that exponential increase in valuation can go on forever

Replying to

and

E.g. see what happens to the math if you assume: a) The person is only using 25% of their money in this particular USD/coin pair and 75% outside of the system b) you only expand out ~10 years

Show replies

Replying to

and

What assumption? That you know what the probability of being correct is?

Replying to

and

What assumptions of kelly do you disagree with?

Replying to

and

1) it's way too conservative for many people 2) it contains strong assumptions about infinitely repeated identical bets that you can't have an adaptive strategy with, none of which describe the real world 3) it's scale-invariant in a way that is inconsistent with its sublinearity