There are two kinds of trading systems. One is the robust type of trading machine, having built in some simple but eternal pieces of trading wisdom. The other is the flexible one, which is able to adapt itself to a special market or stock. This system type could of course also be used to find the holy trading grail and construct a system of type one. Both basic system ideas sound sensible and yet both have severe problems.
Looking for a trading method that is valid for all times and all markets is ambitious, to say the least. What if there is no such trading secret? Fortunately there is and it is indeed somewhat simple. In two words, it is no more than betting on the trend, which is best identified by a chart having a trend line and a current high.
That sounds really simple, and it is a too simple trading system. Either there are some trend trading tricks necessary to prop up this basic trading idea. That would be the practitioners way of trading, the thing the gut trader does. Or the trend method has to be enriched from the mathematical side of trading by finding some trading patterns, some regularity that seems to be always true, regardless of time and market. This is more the way of the trading theoretician, the analyst or the software wizard.
What if we take the trading software approach, eying the adaptive aspect that is used to find the eternal trading advantage and just accept that markets and trading worlds are different?
The answer would then be to have a system that is able to extract current trading regularities, meaning it could be used to find an edge in one market or stock that only works now. Sounds logical, but implicitly we have outspoken the problem here. This trading edge that the software finds works now. More precisely, it has worked until now, in the current market environment.
Will it work tomorrow? Sadly, the answer is almost no. An adaptive trading system that can be trained or optimized by some form of backtesting adapts itself to the history of some price curve and is thus automatically prone to curve fitting. Essentially such a backtesting and self optimizing trading system learns a price chart of the near past. As the backtest shows, it could have done a nearly perfect market timing on that chart. If now confronted with future prices the effectiveness of the optimized system drops nearly to zero.
It is alone this big difference of trading results for the backtest and the gains in reality that should tell everyone that this optimizing method is highly questionable. Even choosing only a few parameters, which backtesting gurus recommend to combat curve fitting, will not change things principally. The system still learns price history with its backtesting algorithm.
This is especially true, because these backtesting algorithms are mostly primitive. They try to optimize what is easily optimizable and that are intervals, periods, frequencies. By doing that these backtesters are inclined to be oscillator systems, which lead to some sort of swing trading in the inner sense, trading the swings. We have here a machine that perfectly learns the frequency of cycles. In other words, it learns the chart of something with its ups and downs.
It is no big miracle that this type of cycle or swing adaption is curve fitting par excellence, no matter how you constrain yourself with the parameters to optimize.
Again, we arrive here at the conclusion that a fusion of these two extremes, the adaptive system and the eternal trading wisdom could be the way to go. The parts that are flexible just have to be defined differently than most backtesting oscillators do it…