If you were to recommend five QF/programming/other books for a guy in my situation, which would they be? Mark Joshi's "concepts" would definitely be on the list.

QuoteOriginally posted by: INFIDELQuoteOriginally posted by: TraderJoeActually, I've read (but not worked through) Hull, as well as a number of papers in SIAM and some put out by the Oxford Uni group. Because of my utter lack of experience in the field, one thing bothers me immensely, especially when reading Hull: what's the relationship of the models to the real world? Hull seems to be full of toy models and I kept wondering how could any of this be realistic? I kept baulking at reading almost any sentence or claim, and spending further time deriving something or doing the exercises, because he seems to be off with the fairies. I would very much like to see a solid connection between model and reality. That's one of the first questions I asked in my face-to-face interview. So what is it exactly that you do? The answer I got was that the group used quantitative models in these places:1) they need to price products. Someone calls up wanting to buy a deriviatives contract, and they need a model to figure out what that contract is worth. If they bid too low, they could lose money. If they bid too high, they could lose business.2) they need to manage risk. Some trader wants to make a trade, and they need to calculate how much money they could lose from the deal so that each trader doesn't exceed their VaR limit.3) they need make sure that the traders have an internally consistent view of the world. The traders are making a number of trades, and part of the purpose of quant models is to make sure that if the trader makes a trade with assumptions X, Y, and Z, that his other trades also make those assumptions.There are also other answers4) You can explicit look for pricing differences between derivatives and model prices. If you find that something should be worth $X, but you can buy it for $X+1, you can make money off of this.5) firm risk management. You need to model that tracks everything the firm does so that if event X happens, the firm does not go under.

QuoteOriginally posted by: energydudeQuoteOriginally posted by: INFIDELYou don't need to talk to experienced quants for this. Most books and research literature do talk about the limitations of the models starting with the falseness of lognormality, volatiility smiles etc. I am surprised that you haven't come across this. Also note that the large number of factors affecting prices is precisely the reason why normal distributions are a somewhat good approximation. It is the finiteness of the factors and their unknowability that is the real problem - I'm being a bit loose here.The way I think about this is that some of the silly assumptions that you see in quantitative finance are similar to the silly assumptions that you see in physics. There is the old joke "assume a spherical cow." You assume a spherical cow, not because cows are spherical, but because you can get a "quick and dirty" answer. Assuming log-normal behavior is like trying to model a pulley system with the assumptions that the rope has no elasticity or that there is no friction. It *isn't* realistic, but it's close enough so that you get some sort of answer. Then you can add more realistic behavior to get better answers.

QuoteOriginally posted by: twofishAssuming log-normal behavior is like trying to model a pulley system with the assumptions that the rope has no elasticity or that there is no friction. It *isn't* realistic, but it's close enough so that you get some sort of answer. Then you can add more realistic behavior to get better answers.The thing is that those sorts of physical models are accurate to zeroth or first order. As long as you know not to quote too many sig fig in your answer, the answers are sensible.Is it the general consensus among quants that in models *as actually used in practice*, market data are:a) always log-normally distributed and you don't really need to fiddle the distribution at allb) always significantly log-normally distributed, and you need to fiddle the distribution (or model predictions) just a little bitc) sometimes, only under special sets of circumstances, significantly log-normally distributed, and it's hard but possible to do something about it d) not even slightly log-normally distributed, and the whole rational modelling enterprise is, at best, like trying to increase the chances of guessing where a roulette wheel will stop based on observing the approximate initial velocity (and location) of the ball? (If you get my drift... i.e. they're still useful. It's like loading the dice in your favor.)(Edited to clear up some English.)

That's the biggest difference with physics for me actually : the models you use in QF actually have an effect on what they're modelling ; if you market your ideas well enough for them to have an impact and deemed appropriate of use by others, then you have a good model.

... I think it'll be very hard to "market" the model that stock returns follow a uniform distribution between two unknown values a and b.(though not impossible )1. You would definitely need support from academia (of the stature of Black, Scholes, Metron etc. )2. Lognormal is bad, untrue and all .... but no one has (yet) figured out a way to make arbitrage profit, high-enough to kill it's usage.(Actually, the model in practice is not pure lognormal, but lognormal giving due to regards to existence of volatility surface ... and further distorted by some market participants using more complex models (jumps and all).I think lognormal is bad ... but not bad-enough to be kicked out (yet).And energydude's wonderful insight only strengthens it's case (.. in addition to the other arguments found in literature).I'd vote for the option (c) sometimes, only under special sets of circumstances, significantly log-normally distributed, and it's hard but possible to do something about it

QuoteOriginally posted by: INFIDELIf you were to recommend five QF/programming/other books for a guy in my situation, which would they be?you've got hull already. I'd have read wilmott instead, but it's maybe overkill to do both, although wilmott (and kwok) are good for the numericsYou likely need to read a little on stochastics, maybe from Neftci: Introduction to the Mathematics of Financial DerivativesYou're likely to be asked stats/probabilty questions in interviews so I'd read a book on that just to refresh.If you're going for risk management jobs, likely Jorion: FRM (=financial risk manager) Handbook will give you an overview of risk management (or some chapters of the better written but multi-author PRM handbook from prmia.org -- wilmott is one of the authors)

QuoteOriginally posted by: INFIDELQuoteOriginally posted by: ppauperQuoteOriginally posted by: INFIDELFor model validation, would I be expected to be so proficient in the theory and practice that I could essentially reproduce the model?not necessarily but depends on the bank.It may just be comparing model outputs to real world dataIf you were to recommend five QF/programming/other books for a guy in my situation, which would they be?Steven E. Shreve : Stochastic Calculus for Finance

Steven E. Shreve : Stochastic Calculus for Finance jedi,It's a very good math book, but INFIDEL want's something related to QF.

When it comes to financial mathematics, I've found that it's a good idea to go through a lot of different books, because each of them have a different way of explaining the topic. The reason I like Hull, is that the concepts in Hull are closer to the one's I'm familiar with, but that also means that I've had to supplement the reading with a lot of different books that exposed me to completely different ways of looking at the problem. (Shreve, Baxter & Rennie, Netifci). I'd also found it useful to look in the Journal of Derivatives to see what's the situation like.As far as the usefulness of the log-normal approximation. My impression is that it provides a useful "baseline" for you to start work from. It is known that real financial products are not log-normal, but you can characterise the real market by how different it is from log-normal (i.e. a real cow is different from a spherical cow because of X, Y, Z).

QuoteOriginally posted by: NSteven E. Shreve : Stochastic Calculus for Finance jedi,It's a very good math book, but INFIDEL want's something related to QF.Well, there are 4 more books one the list.

QuoteOriginally posted by: bhutesI'd vote for the option (c) sometimes, only under special sets of circumstances, significantly log-normally distributed, and it's hard but possible to do something about it I guess that books dealing with what to do about nonlognormality would belong to a "second course" in QF, or at least the material would be shunted towards the end of the standard textbooks. Are there any good books on this (or any good treatments at the end of books) ?

QuoteOriginally posted by: jediWell, there are 4 more books one the list.Which list? Which books?

QuoteOriginally posted by: twofishAs far as the usefulness of the log-normal approximation. My impression is that it provides a useful "baseline" for you to start work from. It is known that real financial products are not log-normal, but you can characterise the real market by how different it is from log-normal (i.e. a real cow is different from a spherical cow because of X, Y, Z).What I'm trying to do is to gauge the extent to which it is realistic. Maybe in the right circumstances, it is very realistic. At the other extreme, though -- in analogy with your metaphor -- you might be trying to answer a question like, "What is the volume of all the angels that fit on a pinhead?" Then it doesn't really help to conceptualize the situation as "a real angel is different from a spherical angel because of X, Y, Z."Call me a killjoy, but I don't believe in angels.

QuoteOriginally posted by: ppauper You're likely to be asked stats/probabilty questions in interviews so I'd read a book on that just to refresh.What sort of depth would you be expected to go into? I hope nothing like: "Please explain Kolmogorov's theory of probability and its significance. The blackboard and chalk are over there, you have 10 minutes."