If I was automating something I probably would not think of scaling out but if I was trading something live and had my mental captal to think about, I would scale out.
We cannot overlook the vital importance of managing emotions in trading. If that means scaling out to be net profitable, so be it. Rigorous testing of mechanical systems taught me that there are optimal profit zones of management, but many other zones of consistent net profits are possible, too. Optimal is what we strive for, but sufficient over the course of time is perfectly acceptable. Chicken & egg debate: theory versus reality, with human emotions playing a pivotal part in the mix
Personally, I don't think that any amount of testing can legitimately invalidate scaling out of a position. Getting out all at once is all-or-nothing behavior. No matter how careful and "scientific" your historical testing may be, the proper interpretation and use of statistics as a forecasting tool does not warrant the kind of confidence that justifies all-or-nothing behavior. It would appear that the people who understand it the least seem to have the most confidence. Just my opinion, of course.
Austin, Although any tests will give one peak (well most any test) it is based on some definition of optimal. Some of those definitions are great on paper and flawed in reality. If you've done the development it seems you'll also know that: 1. most good systems have broad ranges with good results. 2. sometimes you get multiple maxima and if you change the definition of optimal that will usually change the maxima. Scaling out can use multiple maxima to optimize for a broader range of goals. Personally I like to take money home each and every day ... my optimum ... but I also like to participate in the odd home run. Scaling lets me do both.
austinp is absolutely correct that every backtest will come up with a maximum profit at a single exit point. The problem is with curve-fitting and the flaws of it. The optimal profit target will change from day to day and week to week. Whatever is the "optimal" target for a year, will not be the optimal target for the day or week. Therefore, a static "optimal" target can easily be beat by changing profit targets from week to week, day to day, or even hour to hour (depending on your timeframe). It is true that scaling out helps hit a range of targets and will help smooth out P&L over the course of the year. But who cares about that when the point is to make the most money? A second (and possibly more important) advantage to scaling out is that it gives you a real-time gauge of which profit targets are being hit or missed and helps you adjust your profit target range in real-time. Academic backtesting can never compare in reality to the profits that can be achieved via the method I just outlined.