Thank you to Paul and NumberSix for your comments.Regards

Hello, This is an old-topic, but never-ends. Many members in the forums, I think, do fully understand the pitfalls of local vol. and explain in various ways. However, I have some arising questions while reviewing the ideas of the renowned.Seminar works like Dumas&Whaley(1998) conclude that the simpler is better. And I feel some gurus here hate to use the local vol. model and attack its implicit problems with indisputable evidences. Even Paul does an enlightenment campaign that introduce the problem.On the othe hand, in the Derman's lecture note, he also mentions its shortcomings, however, wrote down that with regular update of vol. surface the hedging performance is better than B-S, refering Crepey(2004). He lays emphasis on the "regular update".No matter how it involves various problems, most exotic desks have or use the local vol nowadays, and make dollars.Summing up, what is the best practice? Traders use the local vol. adjusted by themselves(so called the practice) and quants develop their stochastic&jump models in their separate ways. It's a real situation? If not, the local vol. is just for the sake of mark-to-market(lock-in) and P&L depends mostly on traders' speculating or market environment? The one model does out-perfom in a crash market, and the other does in the calm? I wonder traders do really need more sophiscated model.Could you give the light? Please..

The fundamental problem is that stochastic vol models (since they only have a few parameters) don't fit the market very wellGetting greeks from them is difficult/slowSo in general, Stoch vol/Jump models are used to value the trade at inception, maybe do some particular vega hedges then the traders throw it in the "pot" of exotics trades and risk manage the portfolio by local vol.It is hoped that one averages out the errors in the local vol over a portfolio.

Thanks for your reply, spv205.It sounds a good way to use stochastic vol. model in practice. Concensus is that the properly-calibrated stoch. vol. model is better, however, the local vol. is used on a daily basis in investment banks. Right?I've used the local vol model and managed a vega risk by buckets as an interest rate desk does. How can I manage the vega risk with the stoch. vol. model? how can you get the greeks? Shift the imp. vol. surface, recalibrate the parameters, and calculate the difference with them? I don't have any idea how I can hedge the parameter risk with the new greeks. Do you sell/buy the options that have the same amount of parameter sensitivity? This process that brings many calibrations seems to be unstable. I'm not sure even how 'vega' should be defined in the stoch. vol. model.