pricing and stoch calculus - does every quant have to learn them ?

[vol]

I had a look at several lists of recommended reading for aspiring quants.  The usual suspect that everyone recommends are Hull, Joshi, W****tt, etc. I was wondering, for somebody  who wants to work  as a quant in statistical arbitrage/high frequency or automated trading  how usefull are these books ?



Do you really need to know pricing methodologies,  stochastic calculus, for these areas or are you better of reading on time series and econometrics. even for things like volatility arbitrage, or other derivatives related statistical arbitrage



In general, is pricing  theory  usefull  at all if you are going to work in stat arb ? Or is is something that  you still have to show knowledge of at the interviews even if you are not going to use it ?



 



On the same  topic,  should one expect questions from pricing/stoch calc at  interviews with stst arb institutions ?

[Cheng]

I am not into stab art but from what I heard you don't need all that hardcore stoch calc stuff over there. As a matter of fact most of the time you also don't need it in other areas but it is good to have it in the back of one's head.

[FDAXHunter]

What Cheng said.

[kr]

my experience has been that pricing THEORY is not useful at all, but ability to manipulate and interpret the practical stuff is somehow implicit in nearly everything, as it is the back-of-the-envelope quantification of uncertainty of nearly any form. So for instance if you are calculating averages and can't figure out why you are chronically off by sigma^2 / 2, then that would be b/c you missed out on some important stoch calc stuff.



I am a strong supporter of Hull and not too bullish on much of the rest. If you are sitting on the desk and thinking ok, I should figure out what the first approx answer to a certain problem is, then boiling it down to a question of sigma algebras is not going to get you much closer. In fact you will quickly slip into bad real analysis habits of swapping integration order without checking covergence and stuff like that, because you can usually assume away the technicalities (or you realise that a certain structural framework is explosive for some obvious reason).

[dgn2]

I use a lot of back-of-the-envelope calculations relating to geometric Brownian motion to get a quick feel for something, but the vast majority of the time the problems I have are way too messy for stochastic calculus. Pick up the book Computational Statistics Handbook with MATLAB by Wendy L. Martinez and Angel R. Martinez. I think you will find that stat arb is more about quantifying an edge and ensuring that your analysis is robust. Most signal generation that has worked for me has been simple. The layers of control on top of the signal generation tend to be more complex. I do a lot of
Monte Carlo as a way to visualize how different scenarios I can imagine could impact my results.

[Cheng]

vol,



stick to KR's valuable advice. Cutting corners in the right way (ie knowing what error you might produce and estimating it) is crucial in daily work and this can only be done if you know about the foundations.

[vol]

thanks for advice.





One more thing. I have seen several academic papers using stochastic control for portofolio allocation. Basic studies, like the balancing of portofolios between one stock and bonds to maximise expected rate of return under various assumptions. Is anything like that used in practice ? Is it worth going into for a practitioner ?

[Cheng]

Two random thoughts:



- What happens if you choose 200 assets instead of 2 ? Is the method still practical ?

- What do you do about "mispriced" assets like a BBB bond trading with an OAS of +250 (think GM 3 years ago) when the rest of the BBB world trades at +50 ? Your portfolio optimizer will love these assets and fill your portfolio up to the roof with them.

[vol]

Obviously it's not a mature theory that can be efficiently applied in practice but then every theory starts with some toy problems of little practical relevance.



I was wondering if any practitioners see this as a possible future source of capital generation and push for further developments

or is it just one of those short lived theories.

[Cheng]

Just to reiterate what has already been said: if you start with some extremely complex, difficult to handle theory chances are small that you will arrive at something that can be used in practice some day. Unless someone from academia will come with a reasonable approach I fear this theory will not stand the test of time, practitioners usually have more important things on their tables and not enough time.