Most sought-after skills for Quantitative research position
Hi there,
I went to undergrad at a target/semi-target school (top 10 but not ivy) and am currently pursuing a master's in operations research (OR) in an ivy. My plan is to work as a quantitative analyst after graduation.
That being said, I am torn between doing a financial engineering track (taking MBA courses as well as basic OR courses such as optimization and stochastic model) and big data/machine learning track (data analysis and data mining courses). I would love to hear some advice from people who know/currently work in the quantitative finance industry (Hedge fund, Wealth management, etc.)
Thanks in advance!
Majority of real-life quant research is not very mathematically intense, so bear that in mind. Also, depending on what the group does, their requirements will vary. Here is a list of what I looked for when hiring recently: - have at least one programming language that you are very comfortable with - two languages, one prototyping/research and one for power computing are nice - be comfortable with most statistical concepts, it's surprising how many people aren't - basic linear algebra and surrounding data processing ideas (PCA etc) - most importantly, have a brain - that means know what you don't know
Just poking around and I second this. Based on my experience, I'll summarize the difference between the quant concepts you learn in school and those you use on the job:
In school: Assume a stochastic process and derive the quadratic variation. In work: Use R to output the regression coefficients of 1,000 factors driving an asset using data from 2000 to 2018 with prices during tier-1 economic events removed. Summarize the result in two sentences.
I always saw the difference as: one is theoretical, the other is tedious but requires precision
Exactly. A lot of quant finance learned in school is geared towards derivatives pricing. If you are an exotics trader, it actually comes in handy sometimes (not often), but applications to extracting alpha are rather limited.
"- most importantly, have a brain - that means know what you don't know"
+100.
Most seem to be fixated on being smart and intelligent as being what you know, when it is equally knowing what you don't know. h/t Sire.
Could you explain a bit more about the role you hire for? There are a few types of Quants. Almost all seem to have a research component.
I'm mainly looking for quantitative research role. Based on my understanding (which is not much), this role entails practices such as building trading strategies, validating the current model, etc..
For quantitative positions what is preferred more? Python or R?
Given that Python is a General programming language with data analysis libraries while R being almost completely designed for data analysis why is sometimes R given more preference over Python?
A lot of it has to do with the existing codebase and infrastructure. I used to work for a fund that built pretty much everything in R, so I had to learn R. Eventually, I migrated to python, but there are plenty of things possible in R that are not implemented in python quite yet. The choice of programming language mostly depends on what sub-field of quant finance you are trying to get into. If you are aiming for latency-sensetive stuff, you need c++ and there is no way around it. Most general quant finance is either in python, R or some other easy prototyping language. In my group, we mostly build thing in python, with a hefty does of Excel, plus lately we did start doing thing in C++ as we are getting more into lower latency world.
It doesn't really matter much. More important would be that you are comfortable with coding, as linguistic differences between languages stop being obstacles after a certain point in your programming proficiency. If you pick python, and later in your career you have to code in R, it isn't going to be a hassle of learning R syntax (which is in anyway quite straight forward - though CS people will probably kill me for saying this) and adequate documentation of various R libraries also make it easy to become proficient in using specific libraries.
R seems to be a bit niche IMO. You might be better off starting with Python, but if you have a proclivity towards R, then go ahead with it - picking Python if required later isn't tough.
In terms of why R might be given more preference than Python - there isn't really a generic reason why. Perhaps the team using R was using specific libraries? Python can be incredibly painful and frustrating if you are looking to explore and clean data. R is a relative breeze in some cases. But Python can be powerful for other things.
I have a programming background which makes it easier for me to pick up the basics of a language to an certain extent.
But like mentioned above I have been using it for derivatives pricing as a personal project.
I was thinking of picking up Python since I believe data analytics and machine learning will be more prevalent in finance in the future and python allows for both of that with out going into the real nitty gritty.
Also because I have no idea what projects to implement in the languages I know that will be more industry focused as compared to something academic I am hesitant of implementing more derivatives pricers.
PS. Unless it's true HFT you are trying to get into, in that case you better have your IT ducks in a proper linked list
people learn R in schools. School like to pay for licenses for things like RStudio and Matlab. Even Matlab can be seen quoted in some jobs. Python has incorporated the best R has to offer "statistically", but it's much faster and does ML/DL and is used in production systems. go with python. if you know it well, some R shop will just ask you to learn R anyway, which would be a subset. The other way around is a superset. Software Eng don't use R. If you know Python, you're that much closer to having real dev skill ($$), can speak their language and are using arguably the best ML/Quant prog language out there, outside of maybe C++
I would say the big data/machine learning track. Definitely the hot trend going on. Any quantitative asset management conference is pretty much solely big data/ml at this point. I think this will only continue to be the case as the newer datasets get longer histories.
Others are right in that day to day you aren't thinking about crazy math or whatever but it's useful to know. As another poster pointed out, it's arguably much more useful to gain intuition about how to deal with data, do cross-validation, PCA, and other stats techniques
Agree and disagree. Depends how ML is being used. It's certainly not a fad though.
ML methods are fantastic for extracting alpha signals from unconventional datasets, which are then fed into standard quantitative models.
The reason that the ML strategies struggle to get rolled out is because there is still significant pushback from LPs in terms of alpha/factor contribution. When you start using complex machine learning models for actual return forecasting, it becomes difficult if not impossible to attribute your alpha to the different variables that your model is using. Standard quantitative investing approaches give you the ability to do alpha attribution on a variable by variable basis. LPs do not trust this, which is why most of the cutting edge stuff is done at prop shops as well as the top quant funds. And anyway, RenTec is clearly has been using ML so kind of hard to dispute them.
Another major reason as to why machine learning has struggled is because most hedge funds are far behind in technology, particularly non-quant firms that have hired ML people to 'check the box' as something they're looking into. Doing high quality ML research takes both a large amount of computing power and a large amount of clean financial data. People underestimate how difficult it is to clean data such that it is good enough to be put into an ML system. As I'm sure you know, financial data from vendors is often terrible in quality. Regressions tend to be a bit more robust in dealing with shitty data because you can do certain things to minimize the impact and you can interpret them in a more causal fashion. Not the case with ML. ML methods are made to find non-linear patterns and if you give it something incorrect, it may overfit based off garbage data.
There's very little reason to think that the standard regression based approaches are the optimal way of forecasting equities. No reason to think that the many financial datasets that are out there are linearly related. The only reason this is the standard approach is because of the ease of interpret-ability of regression based approaches that are the basis of most active quantitative equity strategies.
..
Hey, thanks so much for the comment. Would you care to elaborate?
Nam unde quia maiores inventore odit aut. Libero consequuntur nam et ex rerum earum. Illo reiciendis temporibus dolor tenetur occaecati itaque cum autem. Consequatur in est perspiciatis sit facere deleniti.
Tenetur porro et quidem id nam autem. Odio est fugit minus voluptatem voluptatum. Commodi debitis officia doloribus quo sed dignissimos. Consequatur hic dolore quidem porro quis.
Ad natus et recusandae voluptates. Accusantium quo in omnis libero hic commodi. Illum autem iure quas et eveniet.
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...