Is stochastic calculus essential for HFT/StatArb quant?

[KurtW]

Hi all,

I'm a newbie and wish to know more about skill sets needed in buy side quants.

My background is master in statistics and PhD in physics. I'm excellent in statistics, R and C++ but clueless in numerical solution of PDE. After doing some internet search, buy side quant doing StatArb or HFT looks suitable for a statistical oriented guy with multi thread C++ programming experience like me.

As the title states, is stochastic calculus and numerical PDE skills essential for StatArb/HFT? If not, do they test your expertise in that area to see if you are smart? I wish to know if my time is better spent learning stochastic calculus, finite difference, MCMC, derivative pricing or elsewhere.

 

[Daniel Duffy]

Numerical solution of PDE and stochastic calculus do not have much overlap.

 

[KurtW]

Numerical solution of PDE and stochastic calculus do not have much overlap.

Thanks for the correction. I thought that numerical methods was used to solve equations in stochastic calculus.
So do you think just a basic understanding of stochastic calculus is enough? By saying basic, I mean an undergrad course level.

 

[Daniel Duffy]

Thanks for the correction. I thought that numerical methods was used to solve equations in stochastic calculus.
So do you think just a basic understanding of stochastic calculus is enough? By saying basic, I mean an undergrad course level.

Actually numerical methods are used for SDE but are very crude compared to PDE.

Numerical solution of SDEs in finance have not got much beyond Euler's method..

 

[Sonny D]

You only need the stochastic for the interviews

 

[Daniel Duffy]

You only need the stochastic for the interviews

Why then so much 'emphasis' on stochastics in education?

 

[Sonny D]

Why then so much 'emphasis' on stochastics in education?

education in Math or Physics? because you need it there. For a quant, there are very few people doing true modeling anymore. Too much capital is required for exotic products, and flow products are simple/finished models. There is some modeling, but not as much as 10 years ago

 

[Daniel Duffy]

education in Math or Physics? because you need it there. For a quant, there are very few people doing true modeling anymore. Too much capital is required for exotic products, and flow products are simple/finished models. There is some modeling, but not as much as 10 years ago

I meant finance.

 

[C S]

Hi all,

I'm a newbie and wish to know more about skill sets needed in buy side quants.

My background is master in statistics and PhD in physics. I'm excellent in statistics, R and C++ but clueless in numerical solution of PDE. After doing some internet search, buy side quant doing StatArb or HFT looks suitable for a statistical oriented guy with multi thread C++ programming experience like me.

As the title states, is stochastic calculus and numerical PDE skills essential for StatArb/HFT? If not, do they test your expertise in that area to see if you are smart? I wish to know if my time is better spent learning stochastic calculus, finite difference, MCMC, derivative pricing or elsewhere.

My general feeling is that stochastic calculus and numerical PDE is not essential for many buy-side systematic trading firms.

As a job hunting strategy, I would not recommend learning something piece-meal just so you can say you know it. Just say "that's not a subject I've studied thoroughly" and the interviewer should move on.

In any real interview where they are trying to actually see if you can add value, they won't care if you didn't memorize some interview questions. They want to see if someone of your skill set and background can contribute. Most small firms will be like that. It's only very large institutions where you will run into canned interview problems that you can answer by studying some quant interview book for a couple months.

 

[Jedison]

For stat arb I don't think Stochastic Calc would be absolutely necessary. Familiarize yourself with a few chapters of Shreve so you have some understanding of basics but a deep understanding of is only necessary if you want to do quant research. You could probably find a niche in data analysis, this field is exploding right now, and not just in finance. I think if you want to write HFT algos you may be fighting an uphill battle against professional programmers. Just my 2 cents.

Edit: I don't mean to say that there isn't a place for you on a HFT team. I would just find a role that is more on the stat/data processing side and less on the super efficient algorithm side.

 

[CasanovaJ]

I meant finance.

Inertia... most MFE faculty are too stubborn to acknowledge that post financial crisis, they should be updating their curricula to focus on things that might actually still be applicable to many of our future careers

Just my (potentially slightly pessimistic) guess.

 

[bigbadwolf]

Inertia... most MFE faculty are too stubborn to acknowledge that post financial crisis, they should be updating their curricula to focus on things that might actually still be applicable to many of our future careers

Just my (potentially slightly pessimistic) guess.

You're correct. In the pre-2008 world, stochastic calc was the jewel in the MFE crown. It was supposed to be the secret esoteric subject that made quants the high priests of finance. But those days are over. Yet MFE faculty still see stochastic calc as the subject that defines the MFE.

 

[C S]

You're correct. In the pre-2008 world, stochastic calc was the jewel in the MFE crown. It was supposed to be the secret esoteric subject that made quants the high priests of finance. But those days are over. Yet MFE faculty still see stochastic calc as the subject that defines the MFE.

It can make sense from their perspective. Some of these programs are run out of math departments. If stoch calc is no longer necessary, then what becomes the "core"? Anything too CS or statistics would suggest the relevant department run the program. Some schools even have competing programs for this very reason.

 

[TehRaio]

It can make sense from their perspective. Some of these programs are run out of math departments. If stoch calc is no longer necessary, then what becomes the "core"? Anything too CS or statistics would suggest the relevant department run the program. Some schools even have competing programs for this very reason.

actually does it make sense to offer multiple programs trying to achieve the same end at the same university?

 

[CasanovaJ]

It can make sense from their perspective. Some of these programs are run out of math departments. If stoch calc is no longer necessary, then what becomes the "core"? Anything too CS or statistics would suggest the relevant department run the program. Some schools even have competing programs for this very reason.

The only job descriptions I've seen that actually appear to require real knowledge of stochastic calculus almost invariably look for "PhDs from top universities," and I should not have had to spend every day in the library this week learning the math behind Asian and Barrier options when the chances of my doing it for a living are virtually nil... The MFE curriculum should be statistics (regular + time series), data analysis, risk analysis, math-heavy finance courses that cover bond math and the basics of financial products, programming, programming, and more programming... stocal should be a single semester course (possibly even an elective) that concisely covers Black Scholes, short rate models, and then calls it a day-- If programs run out of math departments can't handle this, people should choose other programs.

 

[Daniel Duffy]

Inertia... most MFE faculty are too stubborn to acknowledge that post financial crisis, they should be updating their curricula to focus on things that might actually still be applicable to many of our future careers

Just my (potentially slightly pessimistic) guess.

I tend to agree with you. There are two kinds of approach to life in general (which I see/saw in my own neck of the woods):

A. We have a problem, what is the best solution? (engineering)
B. We have a solution/theory that we apply to all problem, irrespective (scientific)

I think B is prevalent even when the facts point to using A approach. Some theory is overrated. I did multiple courses on measure theory during 4-year undergrad (including probability in Hilbert space...) but I have found few applications. That could be my fault. A nice exception is Radon-Nikodym for parameter estimation of SDEs. It's computable.
One last rant Trying to find closed solutions of SDEs.

Personally, I would send all the professors on a QN C++ course :D Like in judo, when you get to 10th black belt you get a white belt and start all over again.

@APalley
@Andy

 

[C S]

I tend to agree with you. There are two kinds of approach to life in general (which I see/saw in my own neck of the woods):

A. We have a problem, what is the best solution? (engineering)
B. We have a solution/theory that we apply to all problem, irrespective (scientific)

I think B is prevalent even when the facts point to using A approach. Some theory is overrated. I did multiple courses on measure theory during 4-year undergrad (including probability in Hilbert space...) but I have found few applications. That could be my fault. A nice exception is Radon-Nikodym for parameter estimation of SDEs. It's computable.
One last rant Trying to find closed solutions of SDEs.

Personally, I would send all the professors on a QN C++ course :D Like in judo, when you get to 10th black belt you get a white belt and start all over again.

@APalley
@Andy

Minor quibble. I think 10th Dan in judo is red.

 

[Daniel Duffy]

Minor quibble. I think 10th Dan in judo is red.

Minor quibble. I think 10th Dan in judo is red.

You are correct.
My recently departed judo trainer Wim Ruska has 8th dan and won TWO gold medals in Munich 1972. He was a legend. RIP.

He had red and white.

I know Tony Sweeney from the Budokwai London and he has 9th dan.

 

[KurtW]

My general feeling is that stochastic calculus and numerical PDE is not essential for many buy-side systematic trading firms.

As a job hunting strategy, I would not recommend learning something piece-meal just so you can say you know it. Just say "that's not a subject I've studied thoroughly" and the interviewer should move on.

In any real interview where they are trying to actually see if you can add value, they won't care if you didn't memorize some interview questions. They want to see if someone of your skill set and background can contribute. Most small firms will be like that. It's only very large institutions where you will run into canned interview problems that you can answer by studying some quant interview book for a couple months.

This is what I thought after doing my reading online. However, most MFE/career guides insists on stochastics, and I get real confused.
Thanks for the insider wisdom here!

 

[KurtW]

For stat arb I don't think Stochastic Calc would be absolutely necessary. Familiarize yourself with a few chapters of Shreve so you have some understanding of basics but a deep understanding of is only necessary if you want to do quant research. You could probably find a niche in data analysis, this field is exploding right now, and not just in finance. I think if you want to write HFT algos you may be fighting an uphill battle against professional programmers. Just my 2 cents.

Edit: I don't mean to say that there isn't a place for you on a HFT team. I would just find a role that is more on the stat/data processing side and less on the super efficient algorithm side.

I think I should follow your advice. My hobby programming experience stands little chance against professionals. Could you provide a link to a statistics oriented quant job so I can spot my missing skills? Thanks a lot!