I read an interesting
article from Robert Pardo: some quotes here without comments. It speaks for itself.
About optmization:
"A lot of people will say, “Let’s try some moving average idea, and optimize and see what we come up with.” They may optimize and find a few models that look really good and completely ignore the fact that most of the rest look pretty bad. I don’t consider optimization to be the way to find if a model is good." About random trading:
"Think about it. If you buy or sell off the flip of a coin, and you know how to manage your risk and your profit, you should be able to make something. I have a former client who had determined, prior to starting with a system, that he would some make money even with a random trade selection. He called it a money management system—I’m not sure that’s what I would call it.
At any rate, you have to ask yourself why people lose so much money trading futures and stocks when by random selection, you should be able to get a fairly even mix of wins and losses. Most people don’t ask that question, but it’s a good question to ask."
About market characteristics and differences:
"Would you also like to see confirmation in other markets?
I’m a little bit different than a lot of people on that count.Let’s say somebody came to me and said “I’ve got a bunch of different models that trade a bunch of different markets, and they’re all different, but they’re all really great.” If I looked at the models and saw they were sound, it wouldn’t bother me a great deal that they wouldn’t work in other markets. Not as long as I felt they were sound in the market they were designed to trade. Individual market characteristics do exist."
About targets and profits:
"With the systems I’ve used over the years, targets have made the models perform less profitably. What you might find is, a model that trades a great deal and is highly accurate could be made more accurate by targets. If you’ve got a model trading 60 percent profitability and it’s a good model, the targets might add 5 to 10 percent to it. So then you’re right seven times out of ten, and for a lot of people, that’s a big difference. They really like being right that often. But that’s another reason people don’t trade well. They’re more interested in being right than they are in just making money."
About accuracy:
"I’ve never really cared about accuracy in a model. A lot of people do, and obviously,the more accurate the better, but I’ve never made that a primary focus in building a model. The models that my current platform started out with were in the area of 45 to 48 percent accurate. They still made a lot of money because they let profits run, and the losses were very manageable. It’s a great irony to me that with no effort to get accuracy, my current models are in the 60 to 65 percent accuracy range."
Systemtrading