Conversation

Replying to

My point is that you're not trying to maximize EV of qwer at all. You are trying to maximize qwer. And whatever qwer is, maximizing expected growth in qwer means that you will ALMOST SURELY maximize qwer itself in the long run

Replying to

and

dude I literally tried to define a made up word because you kept bending words to mean nonstandard things and you responded by disagreeing with the definition of a word I just made up? also: almost surely is not the only thing that matters! outliers matter.

Replying to

I'm not disagreeing with the definition of the word qwer! I am disagreeing that you are trying to maximize the EV of anything.

Replying to

and

Like, fine, to be pedantic: a) if only one sequence will end up happening, there are no interaction effects between them b) So you have a set of possible outcomes and probabilities o_i and p_i c) because of (a), U = sum_i[p_i*f(o_i)] for some function f d) (cont'd)

Replying to

f here is whatever wacky nonlinear function you want. d) define qwer == f(o_i) e) you are trying to maximize the EV of qwer

Replying to

I think we diverge at (c) in the prior tweet, or (e) here. This is in some senses an ideological disagreement, not a mathematical one. You're saying, "hey, I don't know what outcome I'm going to see, so what I'm going to do is weight each by their probability and sum." (ctd.)

Replying to

On the other hand, we're saying, "hey, I don't know what outcome I'm going to see, so what I'm going to do is pick a strategy that will lead to better outcomes for me with probability 1." You're absolutely correct that we miss out on some extremely good outcomes this way.

Replying to

When I was first learning about this, the frustrating part was that I got taught starting in Econ 101 and then as a professional trader that EV is *the* thing that matters. But! It's just a particular number. I now believe maximizing another number fits my preferences better.

Replying to

and

really glad that you're phrasing this clearly! FWIW I think that the paper fails pretty badly to make this prominent. It's a key assumption underpinning it. I do object a bit to "probability 1". That is not a meaningful statement in practice, rather it's "high probability".