Linear regression - recommended by 9/10 quant researchers. The 10th is getting fired for not bringing any reasonable results after half a year of tinkering with neural networks…
Some pople ask me for a list of firms to apply to for HFT QR. Firstly you can find that online or in the Max Dama write-up, secondly here’s *a* list:
Akuna
Citadel Sec
DRW
Flow Traders
Headlands Tech
HRT
IMC
Jane Street
Jump Trading
Maven Sec
Optiver
SIG
Tower Research
Virtu
XTX
My dude what have you not understood from what I’ve just said.
I spend 8-10 hours/day building Sharpe 10 HFT strategies. I have zero desire (and certainly lack of tech and other obstacles) to do so after that in my free time.
Also my employer would have a look at that if I did.
The realization that a number of quants just invest in low cost ETFs on their personal accounts is... worrying.
If these guys just hand it to Fama at the end of the day (when it comes to their own money) then what chance do we have??
Few people know this but financial returns are not normally distributed. Mean-variance optimisation, Sharpe ratio AND the Black-Scholes formula are completely useless and can’t be used to make money.
People be spending so much time with their personal investments these days.
Bro, you can’t out-invest a bad salary.
Maybe if you invested all that time you watch markets into getting better at something and got a better (not only paying) job as a result…
To reduce the dimension of the “problem” let’s say there are three types of role at a (quant/HFT) trading firm:
1.Quant Researcher
2.Quant Trader
3.Quant Developer (Dev)
Some firms mix these terms or call everyone an Analyst (catchall). N1. is sometimes called Strategist.
Factor quants will literally be like: “We sort stocks by how well their previous return predicts the next period return according to R2 (over a moving window), then long the highest decile of those that moved up and short the highest decile of those that moved down.”
Aaand?
Just because I create models that run decisions based on every tick doesn’t mean I know where the fuck is (any) market going tomorrow.
Index funds are very cost efficient, especially with the time they save you.
Some people just cannot take “it depends, …” answers. They want a clear definitive answer and when you start with “it depends, …” and go possibility by possibility, you just completely lose them. They don’t trust you.
Trading is full of “it depends, …” answers, btw.
The allegedly secret trading strategy over which Jane Street Group is suing two former traders and Millennium Management involves options trading in India, lawyers inadvertently revealed at a court hearing: BBG
so the entire Indian stock market is an HFT pump and dump
Popular misconception they’re trying to shill: “You can learn trading all by yourself from public resources, web and books.”
WRONG!
I was trying to go this way a while ago.
I learned more in the first month of joining (mid-tier) prop shop than I did the year before by myself.
That’s a really good honest question.
I’d read and listened to guys like
@macrocephalopod
(threads, podcast w/ Corey)
@therobotjames
(don’t let the shitpostd disturb you)
@AgustinLebron3
(read his book Laws of Trading)
@picotrades
(posting nice brainteasers from the QR space)
@FormulaDeltaOne
I went for these guys as an “Where to start if you want to go into quant trading”.
Seems like that is not the thing to do, so what would you guys then recommend?
So far applying linear regressions to many more statistical problems in HFT has born me so much more fruit than toying with and tweaking some boosted trees or NN architectures on a few selected problems.
BUT… YOU JUST HAVE TO ACCEPT THAT THIS IS THE WAY IT IS.
It’s better to spend weeks-months prepping and interviewing than spending years/a lifetime in a shitty job. Sit the fuck down, open the fucking Heard on the Street and start fucking prepping.
If you want a theoretical foundation in moderately high-frequency market making, check out:
"Algorithmic and High-Frequency Trading" by José Penalva, Sebastian Jaimungal, and Álvaro Cartea (2015)
There are 4 types of things being tested in the interview process directly:
1.Brainteasers + probability&stats
2.Coding (language knowledge + algorithm design - not “algo” as in “strat” lol)
3.Numeracy & “IQ skills”
4.Quick mental math
We are too busy making so much money with C++ as is, we don’t have time to pursue memory safety.
My day-to-day job is writing Python btw so I’m highly qualified to have an opinion on the matter.
Why is no else pursuing memory safety in C++.. I really would like someone to talk shop with. Very frustrating stuff. Disappointed with the C++ community that in the face of this challenge just rolls over and dies.
If you spent 6 months tinkering with ChatGPT I think you should drop the quant in your name, because that implies that you simply cannot have anything to do with real trading.
Although I’m a linear regression connossieur I’d like to brush up on other machine learning techniques. Are there any new up-to-date books about ML that you’d recommend? Even a bit hardcore is ok. 🙏
I have read (basically all) the classics (Elements of Stat Learning et al).
“Yeah I’ve been working on a new market making - I can’t talk about that too much - it’s market making something. (😮) And it’s a very fun one. Had a lot of kind of role in that and it’s part of a I guess this is something …”
🔥Interview
#17
- Stat Arb - 9 April 2024
I chat with Stat Arb (
@quant_arb
) about quant trading, market making, and crypto markets and his experiences founding a fund and working in quant research.
Check it out 👇
Oh no… 🥲 Bye bye, old foe, come back again to read my threads on quant interviewing once you come of age and will want to get a real job at a real firm.
Some ppl here were taken aback by the competitivness of the quant trading job market.
It’s all natural progression. If you’re 9.5/10 and I am 7/10 but I can “gain” 3 points by prepping (while you are complacent with your superior talent) I’m doing that all day, taking your seat.
My POV: hedging = you pay some of your EV for lower variance of PnL. Ergo it’s a movement along the Sharpe x absolute PnL tradeoff curve. Ideally you want to maximize your PnL w/ the constraint that Sharpe drops just enough that you can eat the drawdowns.
@FormulaDeltaOne
@sasuke___420
@QuasiNeutral_
I always tell juniors/interns: Hedging costs money and you should do as little as possible. If a hedge doesn't cost money then it's not a hedge, it's a good trade and you should do more.
If you are a hammer, everything starts to look as a nail.
This is especially true for many quants. Coming from trad ML background? Everything can be solved by a more advanced model. Optimization background? Lemme just set up this objective function…
Why is it always crypto in the DMs, people asking me “hey I’m from exchange xxx do you want to MM for us?”. It’s boring, why is it not e.g. a CME rep asking me to make markets in Lean Hog Futures?
As ppl in TradFi are aware market making doesn’t increase traded volume that much.
This is not an AMA. yet, but what *topics* would you be interested in?
I feel at least a little bit qualified to talk about HFT, systematic trading, predictive models and job interviews. I don’t know jack shit about macro etc.
By meticulously analysing what’s called the cumulative distribution function in options markets I have arrived at a profitable strategy. Buy the bid sell the ask.
Tbh almost any big breakthrough seems counterintuitive, unexpected at first and initially meets with significant skepticism. In other trading firms, in science etc.
Only ex post it is “wow everything falls into place, why didn’t we think of this earlier?”.
So how does Jane Street come up with an options strategy that they don’t understand and don’t believe in? Probably by studying their own loss of efficiency and someone arbing them. $SPX
Yes invenotory management is hedging - reducing variance by paying a bit of EV. If you don’t have alpha, you don’t have EV, so reducing variance is meaningless.
Worry about the quality of your fair price first, then latency, and good inventory mngmt is the cherry on the top.
For all the hopefuls out there. I suggest you stop worrying about what RenTech might or might not be doing. Hanging on every word any firm’s present or former member utters for the press. I assure you, they’re never sharing anything of value.
Everybody’s salty about this one, like they’re writing rocketship navigation systems.
In my industry, quick scripting one can do w/ Python to iterate ideas and find hidden piles of cash is invaluable.
Just bcs you prefer your niche language you don’t have to attack Python.
Good fucking point. Non-normality of residuals being a problem for inference is one of the things that if you say in a quant interview, you prolly not gona make it. Especially if you bridge it into tirade about ML techniques you know.
Usually people bragging about their CFA signals that they think about it as one of the hardest things they’ve done in their life. Which actually tells you a lot…
I've learnt that some people get triggered by my tweets due to the "CFA" next to my name, so after carefully considering and mulling it over for hours I have decided to keep it.
Signs your quant is ngmi
- no sanity checks
- magics away data errors instead of fixing
- instead of pinpointing specific errors, can only point to some complex aggregation
- backtests every symbol and picks the top few
- churns through trading research papers
- not conscientious
This is really really good stuff. The first point should be called “the first rule of trading” and is so often ignored.
Those who trade because they want to feel smart/right will inevitably lose all their money to those who trade because they want to make money. 💸
random thoughts for quant traders
- trading is a business so trade to make money not to be smart, right, admired, or anything else or you will pay for those things
- "The first principle is that you must not fool yourself and you are the easiest person to fool"
- your processes
Hedging being of greater importance to alpha in HFT MM is a strange hill to be willing to die on. Examples:
Alpha shite, hedging superb
-> E[pnl]<0, Var[pnl]=0
(Basically negative infinite Sharpe 👏)
Alpha superb, hedging shite
-> E[pnl]=x>0, Var[pnl]=x/2
I promised a thread on interviewing at (quant) trading firms, here is the first part. I’ve been super busy lately, I’ll continue later.
Everything I say is from my experience, yours can be different so take everything I say with a pinch of salt.
A blast from the past.
Love when I encounter LinkedIn trading influencer talking absolute crap about some basic finance concepts and suddenly Harel (who was an OG here on Twitter) spawns outta nowhere and puts them in their place.
Some folks have been asking which firms to apply to, this is a good start from an OG reference.
Mind you some of the firms have been merged into others, renamed or wound down - the list is old.
39,000 of you follow an account that posts like this (every couple of months or so):
“I’m an experienced HFT market maker”
“I’m an experienced MidFreq portfolio manager”
“I’m an experienced hedge fun manager”
Why would you do that to yourselves?
488,000 of you follow an account that posts analysis like this:
"someone knows something"
"this is very strange"
"Nothing adds up here"
"You can't make this up"
Why do you do this to yourselves?
Even though I work in the TradFi industry and enjoy investing there, as well as building trading algos linked with (regularized!!) linear regression predictive mechanisms, I still think the only way to make serious money is by having an edge and managing your risks well.
Even though I work in the crypto industry and enjoy investing there, as well as building trading algos linked with ML predictive mechanisms, I still think tradfi is the only way to make serious money. Also, I am sentimental.
Here are (symplifying, but not that much) the ONLY resources that you need. Don’t get cute, don’t collect (other) books. Screw all that, just grind it out, it’s about the depth not broadness. You have to make these problems your own and understand them inside out, upside down.
There are 4 types of things being tested in the interview process directly:
1.Brainteasers + probability&stats
2.Coding (language knowledge + algorithm design - not “algo” as in “strat” lol)
3.Numeracy & “IQ skills”
4.Quick mental math
This is spicy. One person who does HFT MM, two who do not. Yet they are convincing
@sasuke___420
about what is useful and what is not. If they were on Twitter only to learn, they would stfu and listen to what sasuke has to say.
Maybe one of these days I’ll do an
AMA AHAIAIIDVANRTMAMODPAODTTPAAIIKAUI
= Ask me anything about HFT and I’ll answer if it doesn’t violate an NDA, reveal too much about me or doesn’t present an opportunity directly translatable to PnL and also if I know and understand it.
Is it just me or have you also been hearing the “buzzword” mid-freq trading everywhere in 2023? Almost as if the HFT arms race ended in temporary division of spheres of influence and now we’re ready to conquer the realms uncharted.
Oh no
@quant_arb
blocked me. Now someone else has to call this
MD/prop trader/HFT MM/whatever he LARPS as today’s
bullshit.
That was the only reason I still followed him. Really hope that not many aspiring quants fall for his mumbo jumbo bullshit…
1. ceph steals rant from me
2. RJ steals from ceph and adds Sydney Sweeney’s boobs, shit goes viral
3. Elon responds with an emoji
🤝
One day Guy Ritchie will shoot a movie about this.
@FormulaDeltaOne
oh lol i thought i was stealing this from ceph, rather than stealing it from ceph stealing it from you.
either way, it's what the lord elon would have wanted. thank you for your service! 🙏❤️
Everybody tells you to work smart, not hard. Many people then don’t work at all if they cannot work smart in some instance.
Whilst the default fallback should be to work hard, if you ever want to accomplish something.
I still love the fact that generations of wannabe quants spent days-months studying HMMs as their holy grail because some interviewee dropped the buzzword in your book. 🙃
The key w/ Renaissance isn’t the brainpower but the collaboration. No one comes close. No silos. No pods. One model. All paid based on how Medallion does.
Jim was as good a manager as he was a quant. Maybe better.
That’s Renaissance’s true secret.
Weekend Twitter is dull, gonna ramble a bit…
In the 90s/00s we had supersonic travel (Concorde), space shuttles etc, now we have blockain virgin bs or nonsense generating “AI”. What the hell happened to our urge for real progress and pushing the boundaries of possible?
Capacity is a serious issue for HFT only approach.
If capacity was the problem for Domeyard (due to pesky investors, I imagine), why didn’t you continue trading but without external money?
Honest researchers acknowledge that there is a replication crisis in finance, among other reasons due to: a) p-hacking; b) backtest overfitting; and c) causal denial.
To learn more, read:
That ain’t happening 😂 unless it’s a bait… No sane person really working in the HFT industry would “walk through code” some random anons or anyone outside their firm for that matter.
Apparently if you say CFA three times, an army of people summons out of thin air that must convince you that the exams are really hard.
Sorry folks if it can be obtained by couple of months studying it ain’t really that hard.
I love the “2 months ago I was in your DMs asking about good MFE programs and basic textbooks on option theory, now I’m an experienced fund manager and successful trader” vibes.
To be fair there are things an independent trader a.k.a. the little guy can do to make money (primary objective) and that is to do low capacity stuff (by definition not exploited by the bigger/better and/or harvest risk premia w/ good risk management.
The Robot guy teaches that.
This is huuuge differentiator. Actively seek mentorship, exact money is secondary at the start of your career. The mentorship usually has higher intrinsic value than your comp.
If you have no edge you start to worry about smart beta. Then u realize your beta wasn’t smart at all and was just poorly timed beta. So you just underperform SPX instead. Like a boss. The lifecycle of an investment manager…
Yep, lower niche engagement and more random public engagement is what makes this feel worse for me as well.
If I’m posting about HFT and some of the people are like “what is HFT”… why do you even follow me? 🤣
feels like the content-selection algorithm is hopelessly broken and I don't want to pay for premium--not bc I really care about twitter getting $84/y but bc I don't want the kind of low-quality attention you get from being boosted to the broad public when you are a niche tweeter
“My strategy performs poorly according to this (for good reason commonly used) metric, e.g. Sharpe), let’s invent this new metric which looks much better…”
- way more common even among finance/trading “professionals” than you’d think
I percieve interviewing @ these firms (quant, HFTs…) and getting an offer as a probabilistic event. Unless you are an absolute genius you won’t get an offer from every single one since processes can have 5+ rounds and are generally hard. You should apply @ as many as possible.
I don’t have experience w/ building these trend portfolios on longer time horizons, but this thread sounds really reasonable.
Also the Sharpe’s sound very realistic from what I’ve been hearing.
~0.2 for single asset
~2 for portfolio of tens-hundreds low-correlation instruments
Couple of thoughs about trend following is probably the most straightforward systematic trading strategy to implement - you're just buying the stuff that's been going up and selling the stuff that's been going down.
So if I were in Chicago, NYC, Bahamas, London, Amsterdam, Singapore, Sydney, Tokyo would you be up to meet in person?
What are some other FinTwit hotspots?
People seem to be very concerned about the order flow toxicity. I might do a thread on why they mostly shouldn’t. Or I also might not. Idk nobody cares here.
Also I’ll probably lock my account for a while.
Good Q. If trading slow, think about how do you wanna execute in prod.
Market orders? Assume you always hit bid/lift offer. Don’t have bests in your data? Add/subtract the halfspread.
Limit orders? Who are you kidding, just use market orders.