Assume you have implemented the Heston, Bates and the SVJJ model and are able to calibrate the models on observed market option prices. You find out the Bates model performs best (in-sample) with a average error of 10.5% when fitted on the price surface. Assume you wish to use these models in practice i.e. start something on your own. Does anyone have some ideas or experience in this field and would be able to give some advice on what one could do? I've always wanted to apply the knowledge I've obtained to real life trading scenarios. Therefore, I would appreciate any relevant input or advice. Best, VolatilityMan

Best fit doesn't imply best model... In my view you cant do much with it unless you're trading exotic and believe that your models capture the vol dynamics (not vol fit) better than the street. And market makers will charge you a healthy spread (because they know from experience none of their models work well - and they will have choice of ~ 20 or more models and a whole team of quants supporting them) so you can say goodbye to most of your edge on the buy side.If you want to start something on your own, keep it simple and look for pricing edge in say spreads or term-structure of vols, rates, divs etc. You wont be able to convert better model into pnl starting on your own.