That's an astute distinction, and maybe I can illustrate with an example using the past two trading days. For the record, I run an overnight system designed solely for ETFs and trade the same ETFs intraday with discretion. The system is a hybrid, combining volatility breakout with mean reversion using Crabel's concepts of range, so it switches back and forth depending upon the expected range.
So, starting with the system, it expected a narrow range day yesterday, so any breakouts were to be faded. Around 2 pm, a breakout was triggered, and the system faded "sold" the breakout as price traversed back below the high of the previously defined range. Since the system had no other signals that day, it held overnight and carried the gap (pure luck but factored into any overnight system). Now, you may be thinking that the system was smart enough to stop-and-reverse on the gap down this morning. Nope -- because historically the majority of large gaps tend to establish a new trend in the gap's direction for at least 3-4 days. The system went long much later only after clearing some key resistance. Thus, although the system made money the past two days, its statistical profile dictated that the gap not be covered on the open, and profit was not maximized.
Now, take the human -- me. QQQQ had been in a prolonged, shallow downtrend near its 50-day MA, but showing good relative strength. This morning at 6 am EST, the NQ's are trading down 30. We're getting the EIA report at 10:30, the employment report tomorrow. What are you going to do if your experience screams go long? Cover of course (half-position), and then watch the price action for some recovery or for an ORB to re-establish a net short position.
Which brings us to system development. Yes, anything can be programmed... anything. But when you're coding special trading cases, you're curve-fitting, and there's no way to establish the robustness of a system with so many degrees of freedom. I completely believe in adaptive systems, but the rules must be relatively simple and work on a walk-forward basis. One way to develop a robust "all-in" system is to include all of the possible signals that you think make sense, but then eliminate those that prove to be superfluous in real-time (reductio ad absurdum). The rule is you can delete code, but rarely can you add code.
I've come to believe that one can develop superb trading skills by running an automated system AND trading with discretion. There's no substitute for watching the market and the patterns develop in real-time. Once your real-time pattern recognition fuses with all those charts you've studied, you're on the path, and then you can use your system framework to guide you intraday. Sometimes the system is better, sometimes you're better. Typically, the discretionary trader blows the doors off the systems trader on NR days and vice versa for WR days.
