wilmott.com

Does anyone know what the interview process for the PhD internship in Securities at GS? I'm doing a CompSci PhD, so while my programming is fine I think my general maths is a bit rusty and financial maths almost non-existent, and I'm just trying to work out what I need to beef up on most before the interviews start. Incidentally, does anyone have any idea how soon after the application deadline (Dec 28th) the interviews will start?Thanks.

There will be far more interviews than you might expect.Don't try learning any maths you don't already have, make sure you are 100% on the maths you say you have.With all due respect, most CS PhDs I encounter abjectly fail to have adequate levels of programming, have you hacked a Linux kernel ? Can you find the median of a data set in linear time ? (I have had 3 CS PhDs who didn't know what that meant Read Hull and Wilmott's intro book, just so you have some idea of the language.What is the job title you are going for ?

what is your PhD topic? Only If you are doing Systems, they would expect you to know a bunch about the kernel and compiler optimization, otherwise it is not required. If it is machine learning, they would expect you to be strong at optimization, matrix theory, concentration bounds, MCMC sampling etc.If it is theory, your math should be very very strong. Typically you end up meeting at least a couple of people who know a lot about your area of research. So expect to be grilled there.You have to be good at algorithms in any case. Revise as much of Cormen as you can. You *should* know C++.Read Joshi's 2 books. Start now. you will know something by the time of your interviews.As far as I know, internship interviews start Feb or so.

QuoteOriginally posted by: veeruthakurwhat is your PhD topic? Only If you are doing Systems, they would expect you to know a bunch about the kernel and compiler optimization, otherwise it is not required. If it is machine learning, they would expect you to be strong at optimization, matrix theory, concentration bounds, MCMC sampling etc.If it is theory, your math should be very very strong. Typically you end up meeting at least a couple of people who know a lot about your area of research. So expect to be grilled there.You have to be good at algorithms in any case. Revise as much of Cormen as you can. You *should* know C++.Read Joshi's 2 books. Start now. you will know something by the time of your interviews.As far as I know, internship interviews start Feb or so.sorry but not sure what you mean by machine learing and "theory"

DCFC - Haven't really hacked the Linux kernel, but I'd consider myself decent enough with general algorithms - I'd find the median in expected O(n) time using a modified quicksort, pivoting on a random element. In terms of the maths I need - I've done PDEs in the past, but no stochastic calculus. Is that likely to be an issue?I'm not sure of the job title - the online application form allowed me to specify 'PhD Internship' and then a division, and shepherded me towards 'Securities' because of having a scientific PhD. I've got fingers crosses that that's put me in the direction of a quant internship. Hopefully the next few places I apply to will allow me to be more specific.veeruthakur - My PhD is on Statistical Machine Translation, so doesn't fit particularly neatly into any of your categories. I'll skim through Cormen again, and see how much of Joshi's books I can get through (I've already started on the first one). Are there any particular parts I should focus on, or just work through it in order? And I presume it'd be beneficial to try and code some of the models?I'm aware that I might not have enough of the bases covered this time around, but figured I'm best off trying to get an internship anyway, and if it all falls through then at least I'll have a decent idea of what I need to learn before I apply for full-time positions this time next year.

Quicksort being O ( N Log(N)) ?It would be nice if you'd worked on some large body of code, the Linux kernel being one example, though you will see elsewhere my views on CS grads who operate only on one level of the system.Yes, the stochastics might hurt, but you can't fix that, so I would instead focus upon the stats that your PhD implies you have.Certainly it's worth hacking through Joshi, and coding stuff up, and Wilmott's FAQ book won't hurt.You ought to make sure you can ace the probability questions in the Brainteasers topic here.There is a scheduling thing here since interviews will be spread over time. and generally they get more specific.

Speaking of quicksort, a useful thread in relation to it and rand() is hereClassic quicksort is O(nlogn) and can degenerate into O(n^2). But STL sort() does not have this problem. BTW how do you get O(n)?

Quicksort is O(n log n), but for median selection (or, for that matter, any element selection), once you've partitioned the data you only need to look at one half of it each time, so you average n + n/2 + n/4 + ... operations, which is O(n).Perhaps I'm misinterpreting, but I get the impression you think CS grads don't have what it takes to be a quant. Is this a general truth? I presume there are some people on here with CS backgrounds who can shed some light... From what I can tell I just need to learn a couple more areas of maths, which I don't see as much of a hurdle (although I admit it may take more time than I have before my first internship interview!).

It is possible to get O(n) based on quicksort algorithm. In original quicksort you choose a pivot then swap data items so that you end up with all items greater than pivot on one side of the array, and all smaller on another side. Then apply the same method recursively on each of the two parts - but in finding the median, you can skip one of them and apply the algorithm only on the relevant part. That is n + n/2 + n/4 + ... + 1 = 2*n, which is linear. At the extreme case, this would have O(n^2) time however.Randomization might be a solution as well. Just choose m (m could be n^(3/4)) random data members from your array, sort them using quicksort, which takes O(mlogm) . As m < n, this step takes less than O(n) time. Then you could take the median from the sorted array. Do another sampling if your guess is wrong.

QuoteOriginally posted by: smiesQuicksort is O(n log n), but for median selection (or, for that matter, any element selection), once you've partitioned the data you only need to look at one half of it each time, so you average n + n/2 + n/4 + ... operations, which is O(n).Perhaps I'm misinterpreting, but I get the impression you think CS grads don't have what it takes to be a quant. Is this a general truth? I presume there are some people on here with CS backgrounds who can shed some light... From what I can tell I just need to learn a couple more areas of maths, which I don't see as much of a hurdle (although I admit it may take more time than I have before my first internship interview!).I would spend several hours to read some important chapters in An Introduction to the Mathematics of Financial Derivatives, Salih Neftci. At least know Ito's formula although you should not put SDE in your CV. If they do ask in the interview, just say you have very basic knowledge in SDE. It happened to me once, and I was lucky to solve several problems in SDE with only limited knowledge in SDE.

QuoteI presume there are some people on here with CS backgrounds who can shed some light.Ok, here is my effort.My own background is not CS as such, but what I have noticed down the years is that, less hard-core stuff is being done now in CS education than say 10, 20, 30 years ago. In particular, the gap between CS and maths on the one hand is becoming greater; On the other hand the programming language(s) (read C++, for example) that industry needs and uses are not being taught in the universities, in general although there is an awareness that something needs to be done.Graph theory, complexity analysis, algebra, state machines, knowledge of h/w is kind of essential.I know of NO university where they teach 'writing maintainable software systems' (maybe CMU?), this is real s/w engineering IMO.When you realise that pricing applications are maths-intensive you need to know the maths. And C++ it would seem.my 2 Euro cent. End of rant.

Cuchulainn - I agree with you about the general trend of CS courses. I'm hoping I've avoided that to an extent by doing what is probably more of a traditional undergraduate CS course (its predecessor was actually the worlds first taught CS course) - it covered all of the topics you mentioned above, and is fairly mathematical.

QuoteOriginally posted by: smiesCuchulainn - I agree with you about the general trend of CS courses. I'm hoping I've avoided that to an extent by doing what is probably more of a traditional undergraduate CS course (its predecessor was actually the worlds first taught CS course) - it covered all of the topics you mentioned above, and is fairly mathematical.smies,was this originally called "Diploma in Numerical Analysis and Automatic Computing"? 1953?

Indeed. The undergraduate teaching developed out of it and started in 1969. It tries to stay close to its origins and remains much more mathematical than CS courses at most other institutions in the UK.

QuoteOriginally posted by: smiesIndeed. The undergraduate teaching developed out of it and started in 1969. It tries to stay close to its origins and remains much more mathematical than CS courses at most other institutions in the UK.Sounds very good. Good luck !