Ideas about new system development, phase 1

Not available any more, sorry, I do not want to disclose it any more.

First of all, welcome to ForexFactory. My name is Ronald Raygun and I'll be your tour guide on this magical journey through the wonders that be.

Just kidding.


Anyway, regarding any sort of crossover system, you're right in that they do in deed work very well during a trending period. Just about any crossover system does during a trending period. What matters more is what happens during a ranging period. I don't see too much in the way of protecting against whipsaw (sideways movement) and would recommend you use something like that.

Regarding your question on the profits, 6,000 pips in 6 months is very good. Many systems here have trouble staying above 0. Here is a trading simulator which can help you see how well the system would have performed in the past.

As for the resilience of the system, a general rule of thumb regarding crossovers, the longer the period, the higher the profitability. I'm not sure if you're any good when it comes to the mathematical aspect of this, but basically, when using crossovers, the two values when put together in every combination, shows a bell curve on the optimization table. Try to predict where the peak will be 1 month from now, 6 months from, now, a year, etc.

Ronald, thank you very much for the reply. Especially for the simulator, now I can test it more thuroughly before programming. I guess the bell curve by your description is the probability density function for normalized distribution and what you are writing to ensure my zero is positive on x asis (to ensure profit mean value is positive). I do not understand much about predicting the peak. If you could explain better (and/or I am wrong with the current understanding), I would appreciate it.

I am macroeconomist, but I am not that good in math (systems of differential equations is where I ended, I prefered statistics much more to math, totally disliking econometrics, but sometimes I have a feeling that the only think I know is a simple regression ), so sometimes I have to ask my friends who teach math advanced courses at universities to help me understand some math. Therefore, you can talk "mathematically" to me .

Also, could you recommend me some protection for sideways whipsaws? I mean to programm them, if I watch it I can reverse position but I want the system to be fully automated. In 90% moments I enter the good direction and position turns positive, I may use tight stoplossses, but would like to use a filter that wold rather delay position closing or speed it up

Is it possible to alter default values in positions handling? The 30 stoploss and profit target? I dont want to use any and would prefer if there was a script for reverse position (and add 5 more speeds between 31 and 32, but you cannot do this I guess), clicking on this one and then just ok. Otherwise it takes too long to backtest this. So far I am fine with visual backtest as I do it, just where the rules are approximately true I enter appropriate arrow and when all are there I delete my indicators to see clearly the arrows and bars and htne just sum the pips in openoffice calc.

Again, thank you very much

N.

PS. I just calculated what it could make on daily oil, and since may 2005 it made more than 24000 pips(I am enclosing nice profit graph), which I consider really nice. But on daily GBPUSD due to whipsaws it made only 7200 pips in two years and only little above 300 pips within last 7 months. GBPUSD hourly profit graph is attached. Maybe you have better idea when checking attached profit graphs. All the graphs the same system.

Well, it's a 3-dimensional bell curve, and in this particular situation, I'll be referring to a basic MA crossover. (Short-term MA crosses up over long-term MA and vice versa)

Next time you're in bed, lay your bed sheet over you, and then draw your knees up so your feet are close to your back while still touching the bed. like so:

http://www.als-mda.org/publications/...ial-sit-up.jpg

The resulting shape that your bedsheet makes is the shape of the 3-d bell curve I referred to earlier.

The x axis is one MA period, the y axis is the other MA period, and the z axis is the total profit. The highest point on this 'bed sheet' changes from time to time, and that was the basis of some of my earliest artificial intelligence EAs. They tried--unsuccessfully--to predict where that peak would be in the near future based on past movement of that peak.

As for protection against whipsaws, the best thing to do is avoid them altogether. Barring that option, you can do two things:

 

1. Use filters to try and remove whipsaws
2. Have a secondary system running to offset the losses that you would otherwise sustain during whipsaw.

There are numerous filters/indicators/methods out there to try and determine whether or not there is a trending period, including:

 

1. Donchian Channels
2. Bollinger Bands
3. MACD
4. AO
5. AC
6. Channel Lines

None of them are nearly as accurate as a well trained human.

If your goal is to make this 100% automated, then I suggest you develop the system to take quick trades during very specific market scenarios as opposed to a system designed to find the trend as much as possible.

Also, what did you mean by the following?

----------------------------------------------------------------------

Is it possible to alter default values in positions handling? The 30 stoploss and profit target? I dont want to use any and would prefer if there was a script for reverse position (and add 5 more speeds between 31 and 32, but you cannot do this I guess), clicking on this one and then just ok. Otherwise it takes too long to backtest this. So far I am fine with visual backtest as I do it, just where the rules are approximately true I enter appropriate arrow and when all are there I delete my indicators to see clearly the arrows and bars and htne just sum the pips in openoffice calc.

Ronald,
I see now what you meant. I did certain tests and found out that settings does not change the bell curve much, it is rather stable. eg. if it would be the ma crossover (in fact, it is not, the crossings are not made by moving averages, but we can say it is since it behaves in a similar way) then

This effect of changing the standard settings one way then it enters sooner and exits in profit sooner, but cannot profit from longer lasting trends with bigger volatility (the system totally ignores sudden temporary plummets or skyrockets in price, inherently it focuses on exhaustion). It seems to be safer, but profits are lower. Changing settings another way enters too lat in a fairly developed trend, say after - bars in trend, which is too late since then they tend to reverse. and also it exits too late, long time after the peak. But I need to do some optimization, maybe there is settings for each asset that will improve final results. My plan is to make an artifical time series from certain futures, stocks and currencies and then optimize it for any market conditions. Maybe I won't have the best performing settings, but the system would be fairly robust. I did play with levenberg-marquardt and other neural networks related algorithms for a year when I found out the neural network can work only for a limited time (like brain, you learn something new and finds out previous information, once new, is totally obsolete). Therefore I do want to avoid fitting, I need to have rational reason for fitting (changing settings), that it should work with different settings for a specific reason (like using mathmematical theory when thinking about volatility). Just trying out different settings if it will make more money is a dead end for systems that struggle staying profitable.

As you wrote about predicting what the bell curve peak would be in a week, month, year, I'd say it is rather a matter of timeframe. I know what you mean, but financial series are not predictable (my experience / I did dissertation on financial crises predictability). So I want to have another approach. Since series tend to trend as economies grow or fall for longer time rather than going sideways, I want to run the script on various timeframes of the same asset, somewhere will be trend and that will make lots of money while others stay around zero. If you have any comments/experience on this, I would welcome it really much since I don't. It is only my reasoning without any experience, maybe it is a dead end.

Also I checked if I run this system on a higher timeframe, then if there is long, I can prohibit shorts on a lower timeframe and profitability would probably increase on longer trends, but I won't catch the turning point trade which is usually profitable for this system (catching the falling knife / I know this should not be done, but why not when it is profitable....I know I am fighting with probailities, but other filters can be added to quickly reverse position back in 2-4 bars, well, this had to be done on the gbpusd h1 maybe three times, which profit graph is also below (this needn't to be done on other asset/timeframe depicted below / so this is other rationale to create the artifical series for optimization. Also I want to create a random walks in excel and test the system on them to test its stability [it is really good thing when working with neural networks since you learn a lot about overfitting when training systems on randomly generated random walks, which financial series are - they are not chaotic, therefore there can be good systems developed in my opinion]).

Thank you very much for the list of indicators, I will try AO, AC and channels since I have no experience with these. I tried donchians last week but they do not help in any way to my system, they cannot decrease risk for it and cannot increase profitability. But definitely thank you for the tip on others, will try them. I tried BB, but they show volatility the way I cannot use., but thanks, now I got an idea how the standard deviation might help me recogniye sideways markets, thanks. MACD is based on moving averages which I hate as lagging indicators . So I won't use it from principle. My system is already lagging entry and sometimes exits, MACD would behave the same at best I guess. There is no rationale it should help. Well, it could, but if you optimize values and fit them, then you can use it as I do my rules, but it won't be stable/robust enough prepared for strange market conditions.

Regarding your question, I meant the default settings in the trading simulator, there are ex4 files, so I cannot edit the default values and am forced to change them each trade. Which takes so long time, that it is almost unusable.

As always, thank you very much, any comments/ideas are welcomed by you and anybody else who has some experience...
N.

I received a PM offering a trend filter in exchange for my indicators. Here is my reply for everybody.
Thanks for the offer, but I don't want to share it right now, I guess I will someday and hope it won't take long months, but please understand I tried to use many systems in previous 4 years, improve them and from the positive beginning and total amazement from an interesting idea it turned to be later just a total loss of time I would not trade in reality.

Maybe if I publish them you all would tell me, wow, this is cool or just the opposite. I want to find the conclusion by myself.

Also, I don't need anybody to programm it. The system is already being programmed for free by people whom I can trust.

If I would give you those indicators (they are free, by traders well known, I found them on forums like forex factory) I'd give you the system in fact. Look, do it like I did, download all the indicators you can, then just go and select only those that look interesting TO YOU (not to somebody else) and combine them. You might find a system that will be yours.

I struggled that I met many profitable systems, but I could not simply trade them because they were too complicated for me, traded way I did not like (like more trades at the same time), had fixed stoplosses or profit targets or there was another thing I did not like about them.
So I decided to play only with indicators I liked and watch them and combined them my way. Truly, what I am using is my idea, I did read about those widely used indicators many pages (even on single page), but nobody had an idea to combine those indicators together.

You can PM me, but please do not ask for the indicators and do not offer programming.

I need advices from experienced traders since I consider myself inexperienced beginner. I have no problem sharing the rules with somebody who really helps me improve the system.
N.