Hi everyone, I've enjoyed this forum a lot, despite it mostly being about IB. Overall career advice here has been really good for the most part. @thebrofessor" suggested last week to do a Q&A so I thought I would give it a shot. A little about me: I went to a target/semi-target and studied math and computer science for undergrad but took several graduate courses and co-authored a few papers in the machine learning realm. Unusual for quants, I do not have a graduate degree. I worked previously at a tier 2 consulting firm as a data scientist after graduating. I have since moved to a quant shop to help lead the effort to integrate machine learning and artificial intelligence into the investment process. Happy to answer any questions WSO might have.
Trending Content
Career Resources
Are you in the top tier quant shop?
How is the working environment and your relationship with the team?
Not on the same tier as two sigma, DE Shaw, etc. but I would say a step below that. I would be interested in applying for one of the top funds in the future though. Also, it's not an HFT shop by any means. We manage pensions, college endowments, etc. Will hold positions for weeks/months at a time. Portfolios are rebalanced using our models periodically. Arguably more of an asset manager than a HF but refer to ourselves ad a HF.
I love the work environment. Head of research gives me types of things he would like to explore but other than that I have a lot of flexibility in terms of what I look into. Most of my day is spent reading new research coming out in machine learning and quantitative finance and thinking of ways to apply it. I will work with my team to determine which approaches would be good to investigate further and then I dive into it. The initial research stages are very independent and hardly collaborative at all. It's only once I've found a promising approach that I pitch my results to others on my team and in portfolio implementation.
What kinds of funds make up that Tier 2 in your opinion?
Did you have sex a non-zero number of times in college?
If I wasn't an athlete, the number would probably have been zero.
Now I'm intrigued. What sport and what division? Do you still like the sport? (My college athlete buddies all ended up hating their sport for various reasons as a result of the experience.) How rough was it managing your time?
Thanks for this, sounds really interesting. Do you try to build systematic strategies, or is the idea to help inform fundamental analysts? How much of your value lies in new data sources vs. new analytic methods (e.g. deep learning or whatever you use)? Can you share your thoughts on how you think these technologies will reshape AM/HF/fintech spaces?
Tier 2 firms: Let me think on this. Not something I've actually thought about since I'm relatively new to the industry.
We build systematic strategies but still have PMs that manually buy/sell. PMs are quants themselves, not fundamental analysts.
New data sources and new analytic methods are two of the big things I'm working on at the moment. Hard to say which is more influential but both are critical at the moment. New analytics methods also follow from new data sources. I am personally trying to push to start using data sources that haven't been around for that long (social media data, etc.). There is some discomfort among management with using data sources that didn't exist 20 years ago due to the current portfolio construction process.
I wrote a lot about this on the quant-mageddon post but I do think that machine learning specifically will reshape the quant space in big ways. My basic argument for this is that the diversity of useful signals that can be extracted from machine learning and AI methods is much greater than traditional quantitative finance method. Machine learning & AI usually requires far less assumptions, which I think is a good thing.
What's your salary and bonus like, and how do you see it progressing at your firm?
What type of quantitative background is needed to be successful with machine learning?
Salary is 110k. Got a relocation/starting bonus of 10k. Performance bonus is supposed to be 30-50% of base. I will find that out soon.
The best background to get a good grasp of machine learning is statistics. When I say statistics, I mean mathematical/graduate level statistics. Although machine learning is often in the computer science department, it really is a statistically driven field. Interpreting results from machine learning algorithms requires a good understanding of type 1 and type 2. If you can learn the stats and math really well, learning the algorithms will come pretty easily.
Edit: In terms of salary progression, I'm not too sure. I'm hoping for a 10k increase in base for next year. Next year's bonus will be more interesting because I will have had a full research life cycle on some of the ideas that I have advanced on the research agenda. If those end up performing well, I think I could see up to 70 to 80% of base bonus.
Thanks for doing this, couple questions...
Do most of the people you work with have phds and do you think you'll go that route?
What coding language would you recommend learning?
It's about 1/3 masters, 1/3 PhDs and 1/3 just undergrad.
Learn math/stats first, programming later. Programming isn't all that difficult to learn. Depends why you want to learn programming in the first place though.
Which specific undergrad or graduate level math course has been the most beneficial for you in your current role?
What kind of projects would you recommend an aspiring quant work in during college?
Were you initially interested in finance then later on developed an interest in machine learning or was it the other way around?
-Most beneficial courses were machine learning, causal inference (favorite course from college) and graduate-level probability theory.
-My experience has been that it's not necessary to have done a lot of finance-specific projects to work as a quant. It's more important that you have good math/stats/programming knowledge and have proven that you have the ability to apply those skills in useful and creative ways.
-I was always interested in working in quantitative finance but didn't get the opportunity to take any financially related courses in college since I was so focused on math and CS. Headhunter reached out to me for my current role and I ended up getting it.
Hello DeepLearning,
Thanks for doing this AMA. I'm in a very similar situation to you, top school non-Masters math graduate and in tier 2 fund in research / development role. So don't mind if I go hard on the questions.
At a minute level, one needs to know clean design patterns, good subclassing and multithreading to keep the flow of trades organized. My guess is that your strategies trade at the top of every hour, which I'm starting to be convinced that you do from reading more of your comments. But are there market inefficiencies at an hourly frequency that can be detected using machine learning? My experience is no.
What exactly constitutes finding a promising approach? Would it be always a backtest result or can it be an analytic result - t-score, Beta plots, ACFs? Put differently, do you get more excited by how there is a pattern between different variables, ie move in CAD aganist move in WTI, or by a backtest return profile of Sharpe 2?
As I mature in my research, I tend to be more focused in the analytical results. I try to find a statistically significant predictor which need not necessarily be readily ported to a trading signal. I prefer these results because they aren't subjected to curve fitting. Once you introduce a backtest, you introduce with it a salvo of data mining biases which shows good in-sample but bad out-of-sample performance. That is, you get negative Sharpe once you trade live.
So when the PMs receive the systematic signal, can they exercise discretion to trade or not trade it? If they can, then this isn't a quant fund. As I heard before, quant is binary, yes or no with nil discretion. Well, if they can't and have to trade regardless, won't the usual set up be a trader, not PM, who trades on the signal.
And this relates to be earlier point. I know a PM from a well established fund who had the exact same set up as you - automated signals, manual execution. And he mentioned that he had to shift to both automated signals and execution to fully realize the edge of the quant strategies.
This reemphasizes my point. You can't be executing quant strategies that sample the market every minute with PMs, or traders, that manually buy and sell. If that's the case, how quant are your strategies when you aren't trading the big moves that can happen anytime within the hour.
Cheers, Nijikon
Okay, I should be more specific. Basic programming isn't that hard. The type of programming required for HFT strategies is extremely difficult. It is accurate that we have a few different strategies and many of the trades are executed at the top of the hour. However, we will hold securities for weeks, sometimes months at a time. We do not do high frequency trading. We have a production team that turns our research python/R code into C++ with multithreading, etc. Converting research into production isn't a big aspect of my role.
It's a backtest approach but backtesting is computationally expensive so we only do that at the later stages. Initially, the analytic results that you mentioned are all important and stuff we use. My specific focus area is machine learning though so the types of models that I build often have slightly different measures of success. With good sampling techniques, avoiding lookahead bias, etc. you can get good out of sample performance. Avoiding overfitting is a legitimate problem but a solvable one with regularization and using training/validation/test sets.
The PMs do not have discretion. I personally think it's kind of stupid that we have PMs and haven't replaced them with an algorithmic trading system but clients are skeptical of replacing PMs with code. If it were up to me I would do exactly as you have suggested and automate the trading.
Hey, I think it's super interesting what you guys do - I've talked to some BlackRock quant people who talked about using algorithms to do deep fundamental research, like tracking consumer activity using social media, or stuff like predicting the short-run economy of China by tracking albedo picked up by satellites from trucks on the road. I've had a pretty long-running fascination with math stats and I've done some CS study in my time so that kind of thing is pretty up my alley!
A couple of questions: I've worked on the E&F investing side of things, and one thing that I've been surprised to see is how people-focused in nature it is; some of our due diligence moves beyond portfolio analysis is boils down to, "do we like the PM, is he a nice guy? Who does he know?" Do you think that deep quant has a place among institutional investors, and how can they take advantage of quant? Not just investing in quant funds but as far as the work they do themselves.
Being that you use quant for fundamental research (and not technical, I assume!) do you think quant has a place in private investing too? Could VC/PE funds use it to predict growth? Are there already such funds doing this?
As DeepLearning has not answered, I'll have an attempt at this.
From what I understand, BlackRock's deep fundamental research is part of the new wave of quant methods which started in the last three years. It really falls under big data, apply quant methods to new non-price datasets unlike what has been done in the last 20 years, which I believe is quant methods applied almost always on price and volume data.
As with anything new starting out, I'm guessing the verdict is still out on whether quant methods has its value in analyzing fundamental data. Let's start with what we know about the big players. BlackRock's quant unit did badly in 2016. Bloomberg had to say:
"At least three of the quant strategies used by BlackRock’s global hedge fund platform have suffered losses greater than 10 percent in the year through November" See: https://www.bloomberg.com/news/articles/2017-01-09/blackrock-quants-sus…
Poin72's big data quant unit is also facing big challenges. We understand that there's a divide between the quants and fundmental money managers. See: https://www.bloomberg.com/news/articles/2017-02-15/point72-shows-how-fi…
Speaking from what I do know, I'll say that quant methods have their value when the data that they are analyzing have some sort of inherent mathematical structure. Consider the most canonical example: the relationship between Canadian dollar and Crude prices. As they are positively correlated, we can employ mathematical tools to predict the correction given a divergence in their prices.
Now, let's shift gears to your big data example, predicting prices movements from social media activity. One can argue that seeing many tweets "iPhone 7 doesn't have earphone jack" or like "Tim Cook is g**" could imply a drop in AAPL. The problem here is in identifying not only what the human reaction is, i.e. buy, sell or hold AAPL, but the rationality and portion of humans that matter. In essence, what we are looking for is out of the sample of humans reading social media, who are the ones that are gonna reflect their sentiment on AAPL's price and is this said group large enough to move the price significantly. There lies the problem in looking at social media data - the rationality of the market is far less understood. The flow of information to prices is obstructed by people who interpreted things in more ways than one.
Contrast this with the market who have CAD exposure. I'm 100% sure that at least 51% of the market will buy CAD when they get to know that Crude prices have increased. This market is certainly more rational when it comes to reflecting information in prices than a social media market, thus the viability in quant method for analysis.
Cheers, Nijikon
Thanks for going in depth with this answer! I've been doing some soul-searching as far as what I want to pursue in Asset Management (which I'm fairly certain I want to stay in), and while I'm familiar with math stats and coding I don't know how far I need to run with it. The firm I'm joining uses stats for the purposes of credit analysis, e.g. they model MBS value for example using statistics about what drives prepayment probability, etc. etc. but I don't know how far that goes. And while my goal is to perhaps one day end up at a long-only fund shop, I don't know what role quant is going to play in that. Is quant inevitable or is its influence contingent on whether allocators believe in it at a critical moment while it's nascent? Is this recent poor performance simply just the firm's inability to work useful signals into an investment thesis, which given time is a problem that can be solved? Or are those signals themselves as useless as the noise they're drawn from? I guess we'll have to wait and see, but maybe you have a more developed opinion on that than I do, haha!
This is interesting. So do you think part of the problem is figuring out a way to quantify the abstract? Like, would it be better if we developed a social media index that could perfectly encapsulate how the internet feels about a stock? Or would that still be rendered useless like you said by not understanding the rationality of the market?
What is your kaggle profile name? what's your current place at Two Sigma Financial Modeling Challenge? would you like to team up for their recruitment competition?
can't believe I missed this AMA...
one thing I am curious about is how quant shops innovate to find the next nut to crack. in my simple brain, I imagine it works like this: the models find some anomaly to exploit, trade it, and then they exit the position. say that anomaly is something like an arb trade between currencies, I doubt that anomaly persists forever, so how do you keep finding new ideas?
I imagine all of the quant shops are thinking along the same lines. do you avoid crowded trades? or, does it not matter?
I too, would like to see a couple examples of "tier 2" firms, because all I'm familiar with is rentech, DE Shaw, Citadel, two sigma, etc.
I would say Acadian Asset Management would be an example of what would be similar to my firm. They're a quant shop but they do not do full on algorithmic trading.
Anomalies don't exist forever, of course. I don't think quant shops are all thinking along the same lines. That was probably the case in 2007 but given the diversity and scale of data sets that are now available, I wouldn't say crowded trades are a problem. With more useful data available, there are more signals to be found.
My team spends about 25 percent of our time reading recent academic papers, whether it be machine learning, AI, statistics, etc. We then determine what aspects of those papers might be applicable for forecasting returns. From there, we collect the relevant data and apply techniques such as principal component analysis in order to determine if the data would actually be useful for forecasting forward returns.
The top quant shops have PhDs that do original research internally, which allows them to take advantage of anomalies for longer.
A quant PM once told me that quant analysts / math specialists may start outpacing / are better suited than (or already are outpacing) equity analysts (fundamental research analysts) on becoming PMs, any general thoughts on this?
Voluptas harum fugiat cum temporibus nulla consequatur. Voluptate excepturi aut id ea porro qui eveniet dignissimos. Et sed libero repellendus et dolorem sint delectus. Quos modi voluptates esse molestias vel aut iste. Sed animi odit qui velit modi praesentium. Laudantium corporis quo temporibus inventore facere ad. Eaque nulla animi eveniet vel molestias.
Aut similique vitae quas repudiandae. Consequuntur et officia qui perferendis aut autem ex. Error modi dignissimos nihil rerum quia nesciunt fugiat. Blanditiis animi repudiandae eligendi quae sint non consequatur.
Est animi tenetur tenetur. Earum et dolorem vel harum et.
Nisi repudiandae blanditiis error est et aut. Aut consequuntur ducimus iste maxime non. Nostrum sit dignissimos nulla odio sequi eius.
See All Comments - 100% Free
WSO depends on everyone being able to pitch in when they know something. Unlock with your email and get bonus: 6 financial modeling lessons free ($199 value)
or Unlock with your social account...
Libero magni minus corporis perspiciatis qui est corporis molestiae. Est ullam sint ipsum amet laborum ut. Et optio blanditiis temporibus officiis qui. Id laudantium illo eos consequatur. Voluptatem nobis recusandae dolores debitis facere laborum quaerat.
In iusto rerum praesentium dolore aut dicta in. Quia sit dolores id. Animi occaecati maxime ut eum et.
Eaque quis ab qui debitis distinctio. Voluptas dolorem tempore dicta rerum. Odio vero veniam magni ea laborum. Architecto ducimus nihil fuga sit repellat ut voluptas pariatur.