ARB8020

ARB8020

12-11-2022

23:11

hello, here's my notes on a Q&A that Max Dama, partner at top tier High Frequency Trading Firm Headlands Technologies hosted

Max graduated UC Berkeley in 2011, joined startup HFT firm Headlands Technologies, birthed from the exodus from Citadel post-2008. he wrote the well-known pdf: and he also has a blog:

on his pdf: ‘i wrote it in college and i didn’t know what was going on, i think its a lot of old information and uninformed’ ‘anything ive written in the past that’s gotten noticed is largely because of a lack of material’ believes his blog is more up-to-date

on majoring in CS/Math/Statistics/Business: quadruple majoring was a waste of time realistically you'll need to know relevant math, stats, and tools for performant code math: linear algebra, numerical analysis stats: distributions, linear models cs: compilers, networking

on HFT event loop/latency: receive exchange data of network packet → decode → add to your ML system → update → buy/sell/do nothing needs to happen before next packet of data comes in (10-20 microseconds) decent cpp code should be able to keep up cont.

ultra low latency maybe makes up 20% of a firm's alpha for the ones that do it at the end of the day most firms are doing different things: not usually as much queue competition the order book updates are pretty slow relative to a CPU

on ML: for HFT, realistically you can calculate a linear combination of some collection of features you have (maybe 100-1000) anything else would have to be mostly stat arb, where you have more time to process and calculate 100s of matrix multiplications or whatever cont.

your data is not iid or stationary linear regression is minimizing mean squared error of your predictions, trading is buying cheap, selling expensive while minimizing costs so one often overlooked thing is figuring out how to map the statistical problem to the trading problem

on keeping up with recent ML advancements: just need to know what % of firms are using say, neural nets, and what % of their alpha comes from it right now neither of those numbers are significant enough for headlands if that number changes, he'll look into it

on strategies that get better when you have more money: not really a thing for the most part acknowledged that economies of scale exist like if you do more volume, you get reduced trading fees, or when you buy computers, buying in bulk gets a discount

on simulation: lots of effort put into it on various fronts simulating network latency is important backtesting, make sure to use a negative orderbook that tracks your own orders, updates when your strategy wouldve executed a take so you don't trade phantom liquidity

on speed of innovation: lots of innovation in HFT infancy, people kind of know how things work now firms are largely different on the niches that they operate in, but most of the stuff is the same

on making trading strategies: what is your value to the market? you're helping disseminate efficient pricing based of supply/demand if gold moves 100bps, should silver move as well? was it a new gold mine that doesn't impact silver, or a change in precious metal demand cont

is the supply/demand change being observed idiosyncratic or does it reflect fundamental change? rich grandma buying tesla and sweeping the book means price more likely to have same theo, get paid to revert the price funds buying tesla off news and sweeping the book is different

on job stress: not really, computer is doing everything 'we're trading a ton of volume in asia right now. i'm over here' not more stressful than any other job for his role

on longer term/macro trades interplay with HFT they don't do a ton of it all the strategies that get run are usually kept separate most firms usually split things up: citadel's macro team doesn't execute through the HFT arm

on tradeoffs of staying at one firm vs jumping around: opportunity cost is pretty big, you don't get to learn, miss out on bonus for length of your noncompete, have to start from the ground up usually at your new firm

on tradeoffs of firm structures: some firms split by functionality citadel's alpha research team -> you research alpha citadel's market impact modeling team -> you model market impact cap on comp varies widely there you do the same thing until you transition cont.

other firms split by team if you're on jump's crypto team, you work on crypto one day you might be doing alpha research another day you might be debugging execution system

on vertical movement/management: largely you get to define how much of that you'd like to do you can stay as a mostly individual contributor, owning pieces of the codebase you can move up and start doing more managing and high level stuff

on being at a small firm: v hard you have to figure out how to do things from the ground up you're not making as much money because the firm isn't making as much money nobody tells you how to find alpha wouldn't recommend joining the headlands that max joined as new grad

on firm rankings: hasn't really changed over time for HFT pretty much just rank by volume at any major exchange was kinda vague, name dropped radix, headlands, jump, citadel, tower, HRT

on stability of HFT industry + more/less openings: there's an economic need for it unlikely to get bigger or smaller generally speaking profit for HFT is volatility x volume some firms expand in profit years, but volatility doesn't stay elevated like that

on market saturation: like mentioned earlier, not really any new firms making waves HFT shook up the paradigm of click trading, doesn't know what's next timekeeping accuracy increase was big, took away one of jump's big trades

there's more detail/less structure in the notes at this link: these are the notes i took live during the talk its super unorganized but maybe might have more you want to see?

all credit goes to Max and the people at the Q&A session, very thankful for what i got to learn there hope you guys enjoyed, let me know if something needs clarification, etc in the raw notes if you choose to subject yourself to that

share if you think that people who follow you might want to see this! would hate for someone who wants to see this to never find it


Follow us on Twitter

to be informed of the latest developments and updates!


You can easily use to @tivitikothread bot for create more readable thread!
Donate 💲

You can keep this app free of charge by supporting 😊

for server charges...