Great question. I hope I can provide a good answer. As you know continuous data is essentially an average of all the contracts at that time. This sounds great until you realize that you could never actually get those prices. Also you will find that that file has no daily range so if you are creating a system that liquidates on stops your stops will certainly lie outside the price reported on that day thus creating a totally unrealistic trading scenario. Personally I built a system using continuous data and stops that worked unbelievably well, it compounded into the trillions of dollars over a 30 year period, but realized too late that this performance was primarily related to compuonding the positive errors of the unrealistic stops. In my opinion just don't use it especially for shorter term systems where these problems would be compounded.
As for what to do, I am still grappling with this problem. I believe you must test each individual contract and carry over the signals of your system if it is still in a trade at the end of the contract period. Also you should be trying to roll your system on to each front month as to capture the most dramatic periods of movement, if this is the aim of your system. I've attached a graph of 1996's Corn contracts to give you an idea of the dramatic differences between contract prices and to illustrate what impact those differences could have on your system.
One last thing, don't get too discouraged about data mistakes. I wasted about 6 weeks using continuous data before I had realized what I had done. At the same time I had just finished reading Mark Ritchie's book God in the Pits, highly recommended, and then reread his interview in New Market Wizards and found that he had wasted six months using continuous data when he built his systems. Hope this wasn't too long of an answer. Best of luck.
