beaver quant

A thread on figuring out how much these dirty perp DEX points are actually worth 🧵 Been having some fun lately so I’m launching perppoints.xyz to value points for my fellow point farmers. Goal: estimate how much $ those points might translate to. 1/7

For my friends who are still using UV and might be a little weary about recent compromises to PyPi packages, stick this in your pyproject.toml. You can let all of those pip users find and report the compromises...

10% discount on vHYPE if you’re willing to take the risk of getting stuck with it. The risk is that the team doesn’t wind down, or no new stakers come in, so you end up stuck with it. I’m happy to hold these until they shut down or some fresh HYPE comes in. Bought some at 0.85.

for it's first weekend, @tradexyz's S&P500 did pretty well volume-wise compared to other markets' first weekends, becoming the biggest equity launch weekend ever on hip-3

Macos tahoe really feel slow, one of the shitest macos update in history

@ScottPh77711570 @turbofish_pk Volume and OI post TGE for many of these projects is going at least -90%. FDV for projects already airdropped is sitting at 3/5x the OI. If you do the math the points are not worth a lot. Better have alpha in determining which project will keep their metrics up

@turbofish_pk Follow up point Any project that gives money for activity which doesn’t increase the value of the project will fail to create value for the token holders

You should expect to be paid for economically useful activity “Farming” is not useful

@nehadesaraju Why?

@idro___ don’t make this mistake 🥲

Rewriting my python infra to rust for learning purposes. This shit is addictive

@BeatzXBT I like the wizard and opaque feeling of Rust

@idro___ you can learn cython you know leave the crabs be

@Cereal_Intern I’m a beginner but rust book and rust by example are great. Just try to code something and figure it out is how I like to learn

@idro___ any good resources for rust?

@klikkiklakki @goatgoatgo97862 I don’t account for everything that happens before I receive the ws message so true latency is higher. Parsing is done with serde_json and communication through lockfree::channel. I have pinned threads for logging, market data feed and prediction computations.

@idro___ @goatgoatgo97862 These numbers sound pretty good! What’s your setup looking like for this? How are you parsing and doing communication with other threads?

@goatgoatgo97862 sadly the rewrite is not finished yet. It's my first time writing rust so not moving really fast here. Biggest latency will probably be due to order signing. But for parsing + formulating a simple prediction I have: - P50: 4us - P99: 14us - P99.9: 57us

@idro___ What's the biggest latency gain so far? The last few steps of executing after triggering a yes isn't covered much.

@ordinaryepsilon Heavy but ordinary polars processing with no weird tricks

@idro___ Why? This seems like a problematic code? E.g. crazy high number of prints/outputs in some loop.

big L for jupyter notebooks again: same script is taking 14min to run in notebooks vs 2.5min in a .py file

@Mtrl_Scientist Same python version (uv in the whole project) + it’s heavy usage of polars

@fearthasea read somewhere (think that it was written by Jeff) that with the taker bump you will trade when your price model disagrees with the MM one. I think it' s a good mental model.

@idro___ i’m curious whether the latency/fee advantage that maker quotes get leads some to quote a bit too comfortably, and whether a good taker signal/latency model could turn that into a workable strat

Back to modeling. Instead of downloaded data I now use locally timestamped data parsed and stored using the same infra as production system. This is a ridge regression only on book features (Binance & Hyperliquid feeds). 1/4

@monkbloc json…

@idro___ What do you use to log data. Is it like a posture or paraquet. Have a hard time building this pipeline

@monkbloc @__qisoka I serialize everything for logging (have a look at msgspec) so types are converted. You can also implement a log data class with a serialize/format method if you need to handle specific things.

@idro___ @__qisoka Do you guys use strict python types? I’m spending a lot of time wrangling with types for things like logging and it feels so verbose. Maybe I’m using python wrong mostly used ts and rust

Started running the system with small size. Signing on HyperLiquid with Python is expensive. Tick to Prediction is ~600us then smash more then 5ms on top of it just to sign the order.

@siddheshrane24 Taker for now

@idro___ what are you trying to model for? MM ?

@dxy2112 Yeah It’s on my to do list

@idro___ So there is somewhere between 0-0.5bps edge here that can generally only be monetized via maker, have you considered trading this on lighter where there’s 0 taker fee? There is a taker speed bump tho but I believe you already account for pessimistic execution

Looking at live data from the system, things are going alright with the realized returns aligning with my prediction. Predictive power is weak but I'm ashamed of what I have done with this model so working on something more predictive that should be better. (This is on BTC)