Developing a Trading System Step by Step

[Murray Ruggiero]

Quote from Capablanca:

So a 3 SMA crossover will be compared to an adaptive donchian channel/turtle breakout to determine which should be the prelimenary setup signal. Okay.

Will the pullback requirement and subsequent entry still be kept or has that been dropped already?

I am not sure yet, we might try it and see what happens. We also might apply other filters and use a fail safe method also.

 

[dis]

Quote from Murray Ruggiero:

Let's now discuss developing our Trend following system. We need to develop the trend detection component. The first one I will discuss is the triple moving average crossover. The triple moving average crossover works different than the dual moving average crossover. The dual moving average crossover has many problems. One example is if I use a short term moving average which is 1/2 the dominant cycle and a long term equal to the dominant cycle , the system is 180% out of phase with the market and I will be buying tops and selling bottoms, except in cases on major trends.

When we use three or more moving averages we don't have this problem.

What evidence do you have that there is a dominant market cycle?

 

[Murray Ruggiero]

In our system we will use the Hilbert Transform to estimate the cycle, you also could use a MEM program like MESA or TradeCycles which I developed for Tradestation or CycleStudio for TradersStudio. These have less lag than Hilbert but there is not code available in the public domain.

A version of Hilbert code was supplied in John Ehlers book
"Rocket Science for Traders "

 

[Xephen]

Murray,

Is CycleStudio available now?

 

[Murray Ruggiero]

It will be available with the Release of 2.0.

 

[trader_rachel]

murray - while you are replenishing good and nice ideas - your cd is long on winding talk but very short on real conclusive content

 

[Murray Ruggiero]

Here are the optimized results for my adaptive channel breakout system. I originally published this method in Futures Magazine in January 1996. I used a MEM addin from Jurik Research in that article because that was before MESA was available for TradeStation and before I developed TradeCycles.

I tested this methodology on a small basket of markets and found that 1.4 times the dominant cycle worked best but the profit between 1.2 and 1.4 differed by less than 10% when I ran this back in 1996.

Last night I ran it on the basket we are using , based on the version of the Hilbert Transform which is included in TradersStudio. I optimized our standard basket from 1/1/80 to 2/21/06. I have included these results in the attachment.

Here is the logic to my simple system.

Sub SimpleAdaptiveChannel(Mult)
Dim MinMove
Dim ChanLen As BarArray
Dim ChanWidth As BarArray
MinMove=GetActiveMinMove()
ChanLen=Min(CInt(CyclePeriod((High+Low)/2,0)*Mult),100)
Buy("ChanBuy",1,Highest(High,ChanLen,0)+MinMove ,Stop,Day)
Sell("ChanSell",1,Lowest(Low,ChanLen,0)-MinMove,Stop,Day)
End Sub

Just like in 1996 , profits differ by less than 10% for Mult in the range of 1.00 to 1.75. We ran in steps of .25 so 1.4 was not included in this run, but it should be in the same range +/ a few percent. I selected 1.25 from this run and this system has done very well over the past ten years.

I have attached the report for the optimization as well as the results for the 1.25 parameter set.

 

[dis]

Quote from Murray Ruggiero:

In our system we will use the Hilbert Transform to estimate the cycle...

Just because data mining may yield a "dominant cycle" does not mean that one exists.

 

[Murray Ruggiero]

That is true , you need to look at the signal to noise ratio which you can get from MEM to see if the cycle is real. What is important is the system I am showing you was developed in 1996!. You can see if has made money for 10 years since release!.

 

[BENG]

That is true , you need to look at the signal to noise ratio which you can get from MEM to see if the cycle is real. What is important is the system I am showing you was developed in 1996!. You can see if has made money for 10 years since release!.

How would the model perform after slippage and commission?