I think you missed my point, which was that software projects sometimes fall behind even if unlimited resources are available. My problems unlike Microsoft's were because I tried to add as many features as possible that were requested by my customers.
We just released version 2.5 of TradersStudio and will begin to post again on various topics. I plan on allocating some time at least 4-5 days a week to posts on ET.
Murray, Did you ever finish this discussion, I have only just started on the path of developing a system. At the moment looking for which software and data provider will best suite my needs. Anyway I was really interested in what your thoughts were on creating better entries and reducing the profits lost, but unfortunately the discussion ended :'(. Any chance of finishing it? Cheers, Inverted
If you have questions ask. Also as you can see this thread got lost and no one even bump it ,so without interest it was hard to invest the time in it. If people are interested I will work on a thread like this again.
Thanks for this. First off you had just decided that the Adaptive Breakout was better than the 3 SMA crossover. You decided this based off the results gathered with no filters at all. Is the best system without filters generally the best system with filters? With regards to the SMA cross over would there be a way to use MEM as you have in the other system to increase performance? On to the lessons that you were left. 1) How to prevent profit give back All of these probably involve filters of some kind. What types of filters do you generally use? 2) How to reduce false entries I have looked at checking further possible areas of resistance, using a longer breakout and checking that the distance between our entry point is of atleast a certain size. 3) How to maximize risk adjusted returns. I take it you are using something like %risk using ATR or a something similar and then just adjusting the %until the profit vs drawdown is maximized? Two last unrelated questions, are there any books that you recommend? Further I have been looking into Fractal Markets, do you have any thoughts regarding this, or links/books that would be helpful Again thanks for doing this. Cheers, James
And one more question: The optimization was done on a basket of markets using 1 lot in each. How do you avoid optimizing only the result of the markets with the largest contract value? For instance, when looking at the code for the SimpleAdaptiveChannel, wouldn't you need to add something to adjust the number of contracts traded, taking into account current volatility, ATR or such? I know TradersStudio has the TradePlan function to adjust the number of contracts and equalize them using risk parameters, but that is applied after all optimization is done. Thanks, Frank
My first pass at optimization is done with a one lot. The next pass I size based on ATR*Bigpointvalue and optimize performance. Ideally I would like to see sets of parameters which performed well in both cases. When optimizing a system making sure that it performs well when the largest winning trades or even the best market is removed. When I developed Trend Harmony, I developed it on 27 markets. Using this large of a basket helps stop one market from overpowering the other. When I first develop a system I leave in markets which perform badly for that reason. You are correct , if you develop a system on a few markets, for example OJ,Corn and the Natural Gas, then Natural gas will overpower the performance. When you have baskets of 20-30 markets that is much harder to have happen, when testing over long periods of time. TradersStudio supports optimization at both the session (system) level and the tradeplan level. The Membership pack contain addins which allow you to analyze optimizations in 3D space as well as apply genetic optimization so that you can analyze large optimization spaces more quickly.
The triple MA is better because the dual is flaw in it's premise. If we are not careful a dual moving average system can be 180 degrees out of phase with the market. This would happen when we have a full cycle lenght moving average minus a 1/2 cycle lenght. The triple moving average system does not have this problem. I have also noticed that a four moving average system does not improve performance enough for the loss in degrees of freedom from the extra parameter. Equity curve giveback is a very hard problem and I have spent a lot of time on it. The problem is any method you use need to be adaptive because each market has different retracements for a move to end. The problem markets are markets like natural gas,coffee which have 2 or more moves which go parabolic each decade. In some markets patterns work well to tell you when you should apply a tight stop to a given market. In other markets these methods do not work. Divergence is also a good tool to use when trying to protect againist giveback. Once you have divergence between MACD for example and price, you need to worry about giveback. The topic of maximize risk adjusted returns is a complex one and I will address it more at a later date along with some more comments about equity curve giveback.
I understand the 2MA vs the 3MA I was just wondering whether the best course was to choose which was better out of the break out and the 3MA before filtering or after. Interesting about the adaptive giveback, can you give an example of this? How would you go about coding divergence in MACD and price using tradersstudio?
There's another dimension to this problem - trade dispersal. A system with a wider distribution of returns is, all things being equal, more risky than another system with a tighter distribution. So, while a system could look good based on your metric the dispersal of trade returns may point to a ticking timebomb of risk. That said, a tighter distribution of returns is not itself proof against drawdown and outlier risk. There is no definitive answer. And this point probably explains why I'm anti Trend Following. Thx D