June was a very busy month, but now that it's over we've finished some very interesting work and would like to share some of it with you in this showcase. What I'm about to talk about represents the future of technical analysis.
Have you ever noticed how every single trading program seems to include copies of SlowK or RSI? Its a funny thing because those tools were designed back when everything had to be calculated and updated by hand. Here we are with computers with clocks running at gigahertz speeds, but we're still stuck on the same old indicators which don't require and weren't designed for computers. Ever try to build a system with RSI or SlowK? You have to be pretty creative, because those tools aren't exact enough to be profitable on a purely overbought/oversold basis.
Let's build a new indicator using modern technology and tools. Using a tool just completed at Market Analytics (and coming!) we'll upgrade our overbought/oversold techniques. The main problem with oscillators is that they blow up when the market trends. What we can do is take a simple SlowK oscillator (good in choppy markets), and combine it with MA_Fractal (good trend detector) to get a new indicator. But the way we'll do it is give both values to the computer, and let it decide how to combine them and what weight to give to each. Using a neural net / genetic algorithm training tool, we found a very good hybrid indicator with the following characteristics on 5 minute S&P data:
Note that we are not building a system, but are merely trying to find an
indicator that performs well. Take a look at the system results shown above.
You'll never see anything like this with those old oscillators. Of course, this
is an in-sample report, so you would never have been able to achieve this. But
this shows the power of this method.
Our software has build a brand new indicator for us based on observed behavior of the underlying market and the indicators we thought were important. There was no training against a shifted target as with commercial neural net products! The shape and style of the indicator were left totally up to the computer. Our only specifications were that this be an overbought/oversold type indicator and that it be profitable.
Here's how things looked the two days following the construction of this indicator:
What sets this oscillator apart from SlowK and the rest is what it did the middle of the 17th. It knew we were in a strong uptrend, so it refused to give any sell signals until the trend was over. Conventional oscillators would likely have given 3-4 false signals during this time period, because they were unaware of the trend. This is extremely valuable behavior. You know what I'm talking about if you've ever tried using classical oscillators in a trending market!
The idea was to build an improved indicator, not a system, but taking a look at the results is still important:
While not as profitable in the unseen period, the indicator still made money when used as a stand alone system. 62% wins and $1500 profit in 2 days trading is not bad for an overbought/oversold indicator, especially considering it was all done in a trending market!
Surround yourself with profitable tools, and its much easier to make a success out of trading because your tools are inherently profitable. Surround yourself with tools that are unprofitable on their own, and you spend all your time trying to filter out bad signals. Hopefully you enjoyed the glimpse into the future! Stay tuned, there are a lot of good things coming just around the corner.
Back to Archives