Sam Babichenko 🇺🇦

I like math. Statistics/Financial Math PhD student at UCSB

[Sam Babichenko 🇺🇦 @SamBabichenko]

Poker is hard. You don't just track the cards, but also manage the beliefs of others, who are doing their best to manage yours. Economists have struggled with this type of problem since the '80s. I solved it for the types of models used in practice.

[Sam Babichenko 🇺🇦 @SamBabichenko]

@mean_field_zane Found mine at a thrift shop for 8 dollars. They thought it was a Halloween costume I guess

[Sam Babichenko 🇺🇦 @SamBabichenko]

@mean_field_zane Who did you take it from? No such thing as a free sleep

[Sam Babichenko 🇺🇦 @SamBabichenko]

@__paleologo With Jane Goodall or the gorillas?

[Sam Babichenko 🇺🇦 @SamBabichenko]

@ben_moll @YaleCowles @econometricsoc My experience is in HA partial information games not HANK, but for me, you have markov transitions for the state and each player's estimates of the shocks. When you do the bellman terms, you get shadow prices for moving the state and shadow prices for moving estimates

[Sam Babichenko 🇺🇦 @SamBabichenko]

@ben_moll @YaleCowles @econometricsoc Now for nonlinear models, you've got the issue the the IRF's aren't deterministic so it's a bit more difficult, but there are a lot of linear phenomena not currently in the literature. I've got a somewhat janky example here: https://t.co/1Fx0hr5HlW

[Sam Babichenko 🇺🇦 @SamBabichenko]

@ben_moll @YaleCowles @econometricsoc Context was for games with partial information (beliefs about beliefs ...)

[Sam Babichenko 🇺🇦 @SamBabichenko]

@ben_moll @YaleCowles @econometricsoc Great presentation. One minor note, it is possible to do dynamic programming with non-markov processes. You can turn any process markov by including the history in the state. Not sure how it works nonlinearly, but I've been able to get it tractable for LQG models

[Sam Babichenko 🇺🇦 @SamBabichenko]

@AgustinLebron3 I prefer flipping stochastic coins actually

[Sam Babichenko 🇺🇦 @SamBabichenko]

@mean_field_zane Is this not equivalent to saying that if two people start from different filtrations and then combine them (or equivalently, that your filtration or a sufficient statistic thereof is measurable to mine and mine to yours), that their estimates of anything would have to match?

[Sam Babichenko 🇺🇦 @SamBabichenko]

@mean_field_zane RIP. At the very least, after one week it’ll be a bad memory rather than a dreadful future

[Sam Babichenko 🇺🇦 @SamBabichenko]

@__paleologo I think a variant of a joke I know applies

[Sam Babichenko 🇺🇦 @SamBabichenko]

@__paleologo When I saw that post, I bought the book and it became one of my favorites (thanks!). I’ll try out this one. Wouldn’t mind if you become a fiction recommendation account

[Sam Babichenko 🇺🇦 @SamBabichenko]

@SandyofCthulhu If it’s a square board (side length s)and it takes a time to cut proportional to the length of the cut, then 15 is correct. First cut has length s, and second has length s/2. If it takes 10 minutes per s, then it’ll take 15 minutes. Definitely not what the teacher was thinking

[Sam Babichenko 🇺🇦 @SamBabichenko]

@quantian1 Yeah, you’re right. I was conflating a short period of high negative jerk then 0 with no third derivative just because it wasn’t a consistent effect

[Sam Babichenko 🇺🇦 @SamBabichenko]

@alexolegimas Yes but some things are scarce due to technology and others are scarce due to politics. AI automation doesn’t make the politically scarce goods cheaper (housing etc) but it does make the technologically scarce goods plentiful. I don’t hear complaining about little access to those

[Sam Babichenko 🇺🇦 @SamBabichenko]

@alz_zyd_ Depends on the area. In my experience doing methods, the hard part is getting adoption

[Sam Babichenko 🇺🇦 @SamBabichenko]

@OliverReganLLC @__paleologo @quantian1 kangaroo. More seriously, it's an asset class where returns dominated by exceptions rather than the norm