I have over 200 strategies that have worked, off and on, over the past 15 years or so. The only one that has worked for longer than a year is this: Buy Home Depot every day except Tuesday near the close and sell it the following trading day about 9:45 Eastern time. This has returned about 20% ROI every year for the past 5 years. You're welcome.
No easy answer cause it's an open question, however I'll give an approach. Predicting probabilities is proved that's more realiable and easy for very short-terms cause certainty ( as some HFT do ), therefore any classic predicting model based on typical timeframes more than 5 minutes are due for eventual failures sometimes, that's the reality and you must deal with it, so the only you can do is pick the best model and code a strict and commom sense money managment rules. The problem with very short timeframes is slipagge, commissions and we, retail, don't have the supervelocity access to prevent fall for unethical HFT "spoofing, quote stuffing...and other wrongdoing" that I hope some day dissapear of markets.
Thank you Steve, I have a serious question for you that's challenging me. I know you are into automated strategies. How did you test these strategies? Manually or program then back test. So, one trading strategy does not work forever? I often hear or read that it often takes testing lots of strategies before finding one that is worth live testing with real money. One webnair trader says he test nearly 100 trading ideas before putting 1 live. If this is the case is nearly impossible to properly manually back test alot of strategies by hand. Appreciate your input Thanks