Introduction & AutoTrading Ideas | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Learning to write code for autotrading in MT4 7 replies

Autotrading with Metatrader while spleeping 24 replies

What are risks with autotrading? 8 replies

Autotrading wih Metatrader 5 replies

AutoTrading with MT4? 0 replies

Attachments: Introduction & AutoTrading Ideas
Exit Attachments

Introduction & AutoTrading Ideas

  • #1
  • First Post: Aug 6, 2007 8:16am Aug 6, 2007 8:16am
Allow myself to introduce myself,

I have been reading these forums at ForexFactory for a few weeks now, and am very impressed with the level of knowledge being shared. I have already found several great ideas floating around that I would not have thought of otherwise. My main interest is in autotrading, and due to this, I thought it would be good to give a short background on myself and start a thread where those interested in autotrading could shared ideas.

My main goal is to LEARN as much as possible, and it seems that by presenting some of my ideas, getting feedback and starting discussions is a great way to do this. I have read some of the other threads on autotrading, and it seems that these use commercial platforms, while I am taking a different approach…

I traded FX manually for about 3 months demo and 6 months live. I ended up at zero profit, due to very poor money management techniques. I was making a lot of small gains, and getting wiped out by over leveraging myself on the wrong ends of trades. Eventually, I decided not to trade manually anymore since I have other full time pursuits. Now I am studying a PhD in computational biology, where I make machine learning models of genetic and regulatory networks. I also studied engineering and worked in software development.

For the last 2 years, I have been working on creating an autotrading system. I have the structure and control system created, but have not gone live due to inadequate demo results. Now I am working on two different models that I will test simultaneously, based on two different forms of machine learning; both look very optimistic. The minimum profit I consider safe to run the system live is 100 pips per month, with a target of 400 per month. Any less than 100 could be dangerous or unreliable in the long term, I believe.

1) I think that it is essential to code everything from scratch. The reason for this is that it is very difficult to control code creation using commercial tools and know exactly what is going on. Even backtesting is tricky if one cannot be certain how it works. I use C++ and R for my work, and highly advise everyone doing this to learn a few languages and control every part of the process. Python is much easier and faster to learn than C++ and Matlab is simpler and more intuitive than R (although expensive, unless you are a student). I should also mention that my strategies probably can’t be coded in EasyLanguage or something similar.

Any comments, or should I continue?

Thanks for you attention,
-Duke
  • #2
  • Aug 6, 2007 8:21am Aug 6, 2007 8:21am
Hmmm... it appears that I accidentally posted in the forum for fully developed systems. My apologies to the moderators for this.

-Duke
  • #3
  • Aug 8, 2007 4:50am Aug 8, 2007 4:50am
Well, perhaps I should elaborate on some things:

Why autotrade? It seems like most people regarding automated/mechanical trading systems as a sort of cop out, a lazy mans alternative to learning a trading strategy and sticking to it. I assure you that have not taken the "quick and easy path", nor am I interested in riches without hard work and effort. The main factors are that I would like to pursue other ventures that trading FX full time would inhibit and that I think the psychological factor in losing money is the hardest obstacle to overcome when trading. Everyone is prone to breaking his rules when things go wrong. Autotrading can *potentially* overcome these limitations. Plus, I just plain like creating systems!

To date, I have put in untold hours creating my testing system, aquiring and formating data, writing testing programs, control programs and analyzing results. It's very difficult, and I am hoping that it will pay off in the long term.

2) It is my belief that ANY trading strategy that is successful can be emulated through a mechanical system. Reading these and other boards, I would guess that 95% of everyone will disagree with this statement, citing other factors, namely intuition, that a system cannot copy. The fact is, all these other factors that are influencing your decision, could be included in the trading system. The main problem is identifying what you are using as all the decision making criterion - from my own trading, I think a HUGE factor in the intuitive decision are the plots of price action that you subconciously look at when deciding if your system's signal is correct. What happens is that your x,y and z indicators are all in alignment to buy. Then you glance at the price over the last day in 5 minute charts (or week in 30 min), and you confirm if you feel that the system is correct. If I am wrong about this, let me know. But no one's system is traded 100% of the time the signals occur, otherwise you could just program it. The main idea is that all these systems that work COULD be programmed.

Anyone care to give the counter argument?

Thanks, -Duke
  • #4
  • Aug 12, 2007 1:26pm Aug 12, 2007 1:26pm
i agree. however, i feel guilty when im having successes with mechanical systems. it literally makes money as you sleep and take a crap. like, at first, i thoguht i was being naive and stupid, but i've seen others do it. it is possible to build a cash cow.

autotrading removes significantly the emotions invovled. also, i can't stay up 24/6 trading, and watching the charts. i get burnt out just watching the fluctuations on my open positions.


genetic algortihm huh? how succesful are your models? i have not programmed any models but im looking into neural networks and gene algorithm, chaos theory.

i doubt models will suddenly become the holy grail, giving out predicitons 100% accuracy. but i think definitely interesting stuff.

lemme know how your gene algorithm goes. feel free to pm me.
{Promotion Removed}
  • #5
  • Aug 12, 2007 2:56pm Aug 12, 2007 2:56pm
Quoting jjk2
i agree. however, i feel guilty when im having successes with mechanical systems. it literally makes money as you sleep and take a crap. like, at first, i thoguht i was being naive and stupid, but i've seen others do it. it is possible to build a cash cow.

autotrading removes significantly the emotions invovled. also, i can't stay up 24/6 trading, and watching the charts. i get burnt out just watching the fluctuations on my open positions.


genetic algortihm huh? how succesful are your models? i have not programmed any models but im looking into neural networks and gene algorithm, chaos theory.

i doubt models will suddenly become the holy grail, giving out predicitons 100% accuracy. but i think definitely interesting stuff.

lemme know how your gene algorithm goes. feel free to pm me.
Hi, thanks for your reply.

I would like to comment on the neural network + genetic algorithm approach. In short: it is dangerous. By that, I mean that it tends to give you good results that do not hold up upon live testing. What it does is give a large number of models, and through the iterative process of optimization, you will end up with a model that will survive 2 or even 3 sets of test data. If you try 10,000 models that are placed on a bell curve of profit, by definition 10 of them will place in the top 10% of the curve 3 times in a row just by chance. Hence, you can get fooled about the real result, which is exactly what happened when I tried this in the beginning of my project.

Neural networks can give good results as they are good at pattern identification, but I do not like them since they are "black boxed". I recommend other forms of machine learning that I will get to later in this thread. But in the end, no method is vastly superior to all others, which is one point I would like to make.

______________________________________________________________________________________________________

3) In my opinion, the system comes down to 3 factors:
1. The idea
2. The quantitative analysis
3. The control system

My belief is that the most important, BY FAR, is 3. The control system. This is why a lot of systems fail. People look at the profits generated by a very crude implementation of their theories without creating a robust and dynamic control structure.

_______________________________________________________________________________________________________


Chaos theory is just a fancy name for nonlinear dynamic systems, that do not reach steady state. They are very useful for modeling natural behaviors, but I do not like them for this type of modeling.

My models work pretty well, but there is room for improvement. Namely, I need them to trade more often in order to fulfill a requirement I have defined. No system can be 100%, but remember that you only need 52% correct trades (assuming a 3 pip spread and 40 pip profit/loss) to make money.

I wrote a pretty long post on Friday which I forgot to take with me from work. I'll put that up tomorrow morning, and hopefull get some comments.

Cheers,
-Duke
  • #6
  • Aug 12, 2007 8:18pm Aug 12, 2007 8:18pm
I have been thinking about starting a thread like this my self, but I will jump on this one...

I am an computer science/math student and some time ago I persuaded my supervisor to let me and one of my friends do a master degree in "Computational Finance" - which in this case translates into automated forex trading. So now I trade forex and get study credit at the same time - hehe..

The first project we made was to build an "agent"* based on machine learning with zero knowledge of the forex market prior to what it learned by itself. The methods we used was a special version of Reinforcement Learning, genetic evolved Neural Networks and finally Support Vector Machines.

The way we approached the subject mirrored our own understanding of the market - which was equal to zero, none of us had any prior experience with the forex market or trading in general. For quite a while I was even wondering about why there where this strange gab in the market every first Friday in the month - hahaha...

The results of the first project were not at all promising. We were inspired by articles written by university researchers with the same lack of market knowledge as our self, and the outcome was pretty much as one could expect. But it sparked an increasing interest in how this market actually works. So I started doing some real research...

What I have come to believe now is that in order to make an automatized trading system it is crucial to know as much as possible about how the forex market functions. Every little detail matters in the quest of finding a statistical edge. The project we are working on now tries to exploit the
knowledge of the forex market that we have been building up combined with various machine learning techniques (all though it might end up more as an expert system).

To comment on some of your views:

1) I completely agree with you that it is important to code everything from scratch. I would never rely on something like Metatrader for back testing for example. If I were to use a program I did not program my self then it had to be open source. We program in C# because it makes it a bit more easy to implement certain features like data visualization, but thats of course a personal opinion. R is nice though it tends to have some problems with large amounts of data.

2) Regarding that any trading strategy can be emulated by a computer I agree in principle. But it demands that the computer has access to the same information as the trader. If the trader base her strategy on watching CNBC (for whatever reason) then you have to be a bit creative in replicating that strategy...(I'm not saying its impossible...).

3) I would add to the list:

4. Detailed knowledge and understanding of the market in question.
5. Reliable(!) data for building the system/model/agent.
(not necessarily in that order)

Damn - its 3 in the morning. Gotta check out the bed.

*) In machine learning a program which is able to react autonomously is often referred to as an agent.
  • #7
  • Aug 12, 2007 8:44pm Aug 12, 2007 8:44pm
i just discovered a MBA thesis on technical trading written 10 years ago. lo and behold, the system has 80~90% success rate, easily picking out top and bottoms. from the backtesting it has more than 30 years of data from commodities market. 5000 -> a million in less than 4 years "theoretically" proposed by the system.

and guess what? the system is not based on complex understanding of chaos math, neural, genetic algorithms, but simply the work of trial and error of various indicators combined. dont think if i know the extra page on the back of my math textbook, i will create abetter more profitable system. i thought like this before....i was gravenly disappointed.


i am implementing this to the forex market. however, i can't publish anything here, since i dont have permission from the author, and the mba thesis book is a property of the university.

i guess my point is, dont rely on complex models too much. they are not teh sole answer. its certainly helpful if u succeed in implementing it. also, the market is made by people, and consists of people. we are not that rational. how can u take all that into account when implementing models? i tend to think of the market as an organic being, breathing on its own, its alive, has emotions. i try set my system to exploit and make money on the tiniest mood swings of its participants.


but computational finance, im definitely interested in unfortunately my schools not that big, we only have a financial engineering, and some asset allocation masters degree, and some risk mangement crap. i think some of the stuff is figuring out prices on derivatives....i think those are boring crap.
{Promotion Removed}
  • #8
  • Aug 13, 2007 3:01am Aug 13, 2007 3:01am
Quoting BlackMage
I have been thinking about starting a thread like this my self, but I will jump on this one...

The first project we made was to build an "agent"* based on machine learning with zero knowledge of the forex market prior to what it learned by itself. The methods we used was a special version of Reinforcement Learning, genetic evolved Neural Networks and finally Support Vector Machines.

What I have come to believe now is that in order to make an automatized trading system it is crucial to know as much as possible about how the forex market functions. Every little detail matters in the quest of finding a statistical edge. The project we are working on now tries to exploit the
knowledge of the forex market that we have been building up combined with various machine learning techniques (all though it might end up more as an expert system).

4. Detailed knowledge and understanding of the market in question.
5. Reliable(!) data for building the system/model/agent.
(not necessarily in that order)
Hello Blackmage!

I'm glad you dropped by, as one intention of this thread is to attract people exactly like yourself!

It's funny that you mention the techniques that you use. I had not yet discussed my methods because they are a bit specific to the machine learning world, but indeed I am basing my models on Support Vector Machines as well as MARS (Multivariate Adaptive Regression Splines), using Reinforcement Learning to build up the control structure.

I agree with you that every detail matters, and that the statistical edge you get is paramount. What I have found with my research is that there are several key points I have discovered that are extremely counterintuitive, and these helped me solve a few key problems in the development phase.

I mention the 3 main factors (which we have now expanded to 5!). I should add that 1) The idea is also extremely important. It doesn't matter what approach you take, how valid your model is if you don't have a proper idea of how the system works. Trying to mine out data for decision making has been futile, in my experience. This is a very key point I would like to make!! The idea for the system has to be predetermined as opposed to trying to create it on-the-fly through data mining and reinforcement learning - because, how exactly are you going to reinforce the model? That's the important "thought experiment"...

It sounds like you are taking a machine learning approach to behavioral finance to examining how the markets react. That is very cool stuff, and I look forward to hearing more of your thoughts and ideas!

Regards,
-Duke
  • #9
  • Aug 13, 2007 3:20am Aug 13, 2007 3:20am
Quoting jjk2
i just discovered a MBA thesis on technical trading written 10 years ago. lo and behold, the system has 80~90% success rate, easily picking out top and bottoms. from the backtesting it has more than 30 years of data from commodities market. 5000 -> a million in less than 4 years "theoretically" proposed by the system.

and guess what? the system is not based on complex understanding of chaos math, neural, genetic algorithms, but simply the work of trial and error of various indicators combined. dont think if i know the extra page on the back of my math textbook, i will create abetter more profitable system. i thought like this before....i was gravenly disappointed.

i guess my point is, dont rely on complex models too much. they are not teh sole answer. its certainly helpful if u succeed in implementing it. also, the market is made by people, and consists of people. we are not that rational. how can u take all that into account when implementing models? i tend to think of the market as an organic being, breathing on its own, its alive, has emotions. i try set my system to exploit and make money on the tiniest mood swings of its participants.
Hi jjk2!

Well, it's impossible to say exactly what is going on without seeing the thesis, but it is farly easy to fit the data to pick tops and bottoms. This is especially true if you ignore the timescale, so that each top and bottom is +1000 pips apart, and that maximum downdraw might be -600 pips. In reality, you could never survive such swings, even with enormous profits.

Any system is valid if and only if it works during forward testing. It is very common that a backtested works well, only to fail during forward testing because 1) the model fit to previous data instead of finding generalized conditions, and therefore fails or 2) backtesting accidentally looked 1 period into the future, which skews results badly.

You are right about the complexity of the markets and the unpredictability of trader mentality. But remember also that price is king. All the ebbs and flows of the market, the "animal spirits" if you will, are reflected in the price and the volume! So this data gives us all the information we need to create a model that will trade these sentiments.

Do we need to be 100% accurate? No.
Do we need to know why things are moving as they do? Debatable, but my view is that no, we do not.

We do need to be able to trade with greater than 52% accuracy (or less with higher average gain than loss). The proof is in the pudding with these systems - they are valid when they work!

Take care,
-Duke
  • #10
  • Edited 8:28am Aug 13, 2007 4:20am | Edited 8:28am
Quoting duketoga
2) It is my belief that ANY trading strategy that is successful can be emulated through a mechanical system. Reading these and other boards, I would guess that 95% of everyone will disagree with this statement, citing other factors, namely intuition, that a system cannot copy. The fact is, all these other factors that are influencing your decision, could be included in the trading system. The main problem is identifying what you are using as all the decision making criterion - from my own trading, I think a HUGE factor in the intuitive decision are the plots of price action that you subconciously look at when deciding if your system's signal is correct. What happens is that your x,y and z indicators are all in alignment to buy. Then you glance at the price over the last day in 5 minute charts (or week in 30 min), and you confirm if you feel that the system is correct. If I am wrong about this, let me know. But no one's system is traded 100% of the time the signals occur, otherwise you could just program it. The main idea is that all these systems that work COULD be programmed.
I do not agree with that point. See Boxingislife sexystochastic strategy. I use it with great success (i'm over 100% in 1.5 month of trading). I have worked on an EA with other ppl but this strategy is so visual that it's not programmable.

There is so many parameters to take into account that an EA picking up all those parameters would probably finaly not take the trade.

edit : i screwed up my % calculation ^^ it's 80% dont know why i though it was 100% maybe i though my starting account was smaller that what i really put in
  • #11
  • Aug 13, 2007 4:41am Aug 13, 2007 4:41am
Quoting melpheos
I do not agree with that point. See Boxingislife sexystochastic strategy. I use it with great success (i'm over 100% in 1.5 month of trading). I have worked on an EA with other ppl but this strategy is so visual that it's not programmable.

There is so many parameters to take into account that an EA picking up all those parameters would probably finaly not take the trade.
Hi Melpheos,

I think I should clarify this point. I don't mean that every trading system can be programmed easily or to great accuracy at the present. I mean that it is theoretically possible to emulate through programming any set of decision making criteria, although this quest may take years of research.

There are definitely a lot of subtle factors that successful traders use that even they probably aren't conciously aware of. Now, for a system such as the one that you are programming, I do believe that is possible at the present, if you continue your work.

Which are the parameters that are "one too many" and are blocking the trade? If you remove them, are there too many trades? The main problem with parameter based system is this: if you list out, say 5 main paramaters (either numerical values or binary), for the trade, you could then look at each potential trade and see what the parameters are. You would run through a whole year and then tally the results. Then you could categorize each potential trade as yes and no. But when you then look at the parameters for each category, they don't show an obvious set of differences. WTF??

The answer is that you need to use soft decision making processes instead of a rigid set of rules or linear regression. This is where machine learning methods come into play. Also, this is why I recommend to use R, Matlab or C++. They are free (except Matlab) and toolboxes for machine learning methods such as neural networks, support vector machines, MARS, etc. are all made by academics and can be freely downloaded.

Hell, you can easily program some decision trees yourself to get a feel for things. I do recommend that you try to learn some of these methods, and while they might not mimic an existing strategy exactly, they might give you insight into creating a new, similar trading method.

Good luck,
-Duke
  • #12
  • Aug 13, 2007 4:45am Aug 13, 2007 4:45am
Quoting duketoga
Hi Melpheos,

I think I should clarify this point. I don't mean that every trading system can be programmed easily or to great accuracy at the present. I mean that it is theoretically possible to emulate through programming any set of decision making criteria, although this quest may take years of research.

There are definitely a lot of subtle factors that successful traders use that even they probably aren't conciously aware of. Now, for a system such as the one that you are programming, I do believe that is possible at the present, if you continue your work.

Which are the parameters that are "one too many" and are blocking the trade? If you remove them, are there too many trades? The main problem with parameter based system is this: if you list out, say 5 main paramaters (either numerical values or binary), for the trade, you could then look at each potential trade and see what the parameters are. You would run through a whole year and then tally the results. Then you could categorize each potential trade as yes and no. But when you then look at the parameters for each category, they don't show an obvious set of differences. WTF??

The answer is that you need to use soft decision making processes instead of a rigid set of rules or linear regression. This is where machine learning methods come into play. Also, this is why I recommend to use R, Matlab or C++. They are free (except Matlab) and toolboxes for machine learning methods such as neural networks, support vector machines, MARS, etc. are all made by academics and can be freely downloaded.

Hell, you can easily program some decision trees yourself to get a feel for things. I do recommend that you try to learn some of these methods, and while they might not mimic an existing strategy exactly, they might give you insight into creating a new, similar trading method.

Good luck,
-Duke
I think i dont have the knowledge level required for such a work :-/ Neural networks etc doesn't seems to be something i could handle without going back to school...
  • #13
  • Aug 13, 2007 6:12am Aug 13, 2007 6:12am
Glad to see that some people are interested in this!!

It seems like many autotraders here are using EAs via the MT4 platform. Perhaps the nature of my work is rather different from those who aren’t going to code themselves from scratch. But I assure you, learning this will pay off in the long term. I was initially able to test a huge amount of strategies to see what worked and what didn’t by running scripts while I went to work. Of course, that only taught me that I need more sophisticated methods J.

When I get my live system running, I might start a journal thread to show its progress.

4) However, I will NEVER, EVER offer to sell it or have a signal service. I see that here at FF you have had some problems with people coming out of the blue, showing a nice system, and then springing their angle. (In fact there is one incredibly fraudulent journal thread that is laughably fake, and happens to use a commercial signal, hmm…). This will NOT happen with me. As everyone knows, any profitable automated system would make more money by trading it than by selling the signals. 400 pips consistently per month on 10 lots of 100k is a LOT of money.

5) Myth #1: Mechanical systems may work initially, but will degrade over time. This is a common statement that I do not believe is true. I cannot verify it with documentation as I haven’t gone live yet. However, it seems to me that the reason for this widespread belief is that people test a lot of autotrading systems, which I mention in an earlier post “Regarding NN and GA”. The system on the right side of the bell curve 3 times in a row looks super, so you decide to run it live. Now even this has a chance to make money, and if it does initially, you get the idea that it works but degrades over time. In my experience, a system either works or it doesn’t, but you need a large number of trades to see the statistics clearly.

The other common theory is that you are capturing some parameters of the market that are valid for the current time only. This I do not believe in, as the goal of any mechanical system is to react to all long-term conditions (see point 6 below). Many say this isn’t possible, but of course it is… you just need a whole lot of theory behind it and a solid control system, and not just 3-5 technical indicators. (I may get killed on this statement due to “changing market geometry”, but the point is that the system should be robust enough to handle changing conditions.)

6) Myth #2: Mechanical systems work well in trending markets, but break down during ranging markets. This is certainly based on the fact that all systems that use technical indicators are inherently lagging. By definition they have to be, since you can’t have information about the price until the price is given. Therefore, systems using MAs, MACD, etc. will all be prone to whipsaws in the time frame you are looking at in a ranging market. I realized this very soon in my research when I tried to duplicate conditions present in huge upswings, but then realized that trading them EVERY time results in 2-3 bad trades for every 1 big gain.

They key here is to understand that you can trade by examining smaller time periods to capture these moves within a ranging market. Of course then the effect of the spread is increased. However, the possibility of reversing a bad trade and exiting the whipsaw is that you will lose the spread more often, but not the pips on fast reversals where you are lagging.

7) Myth #3: Automated systems are risky to trade. Again, I have no idea where this would originate. It is VERY easy to program a system to have very simple safety controls, such as having only 1 lot open at a time, S/L for all transactions, etc. Any system that buys and sells randomly will, over a long time, average 50% and just lose an average of the spread (say $30 per trade on a 100k account). Unless you are tragically unlucky, this should take a while to blow out your account, assuming you don’t trade too frequently - in any case you should notice the performance problem. The pundits will mention consecutive losses and downdraw; however, as I mentioned, a system should reverse in choppy market conditions to prevent large consecutive losses (whereas if it only “cherry picks” its trades, it can and will eventually have 8+ losses in a row). You system should be tested thoroughly on demo trading as well, and any system not bringing in reliable profits should be shelved.

8) LL: leading limit. Well, everyone knows about a trailing stop, so I like the idea of a leading limit (apologies if this has already been introduced). The idea is that when you open a trade, you perhaps have a 30 pip trailing stop and a 100 pip limit. When you have moved some 30 pips profit, your stop is now at break even and your limit is 70 pips additional profit. What if some random spike or news comes out and moves the market 40 pips either way? Well, you have the potential to get screwed big time if you can get stopped out but not take profit. A leading limit will be placed 30 pips ahead of the trade, with maximum limit being set at 100 pips. When your trade moves some number of pips, you replace the limit 30 pips away, so that you are always placed equidistantly between the S and L. Any news spikes will be 50/50 for profit or loss, and therefore should equal out over time. When you get to your absolute limit, you just leave it there (or preferably, your system recalculates that trade and determines how to proceed).

-Duke
  • #14
  • Aug 13, 2007 6:39am Aug 13, 2007 6:39am
I trade using a fully automatic system. It actually trades four currency pairs using two different strategies via a broker API. The 'system' runs independantly on a server.

I am certainly no great heroic trader nor a fantastic programmer but I certainly think it is worth setting things up properly to run completely mechanically. The biggest problem for those that like trading and the adrenalin rush that they might get is that this type of trading is completely boring. I do not even have to look at anything unless I receive an email from the 'system' saying something is wrong and that is rare. I then check out the end of month result at the end of each month.

I support a number of points made on this thread. I always find it strange though that discussions like this are generally amongst the most unread. Must be the boredom thing. I think you guys here are on the right track though.
  • #15
  • Aug 13, 2007 6:49am Aug 13, 2007 6:49am
Duke,

I find your topic of large interest to me as well. I developed an EA called Flyer! which is available in the forums. While it is of course based in the MT4 platform, I like the concept of being able to freely program outside of the platform where much more complicated and faster systems could be implemented. C++ is such a powerful language and can be used to implement virtually anything, especially when it is interfaced with other tools.

My educational background is Computer Science as well and I continue to work on some of my own projects privately at this time. As to trading experience, I have been trading forex now for about 2 years and the equities market for a lot longer.

As to the theory of creating trading programs... I think it is certainly a workable concept. I know there are programs out there that do have successful returns. Most of those are kept in high secrecy though and are not discussed much (otherwise everybody would have one and the markets would counteradjust). But, again we look at the success of computer algorithms to understand other human related concepts and find a good degree of exceptional performance from those (chess playing programs, voice recognition software, etc, etc.). Why not a computerized trader? Certainly I feel it is possible with a degree of accuracy that can be equivalent or if not better than 90% of most human traders. Remember, the whole principle of trading is to buy low and sell high. That is our fundamental goal. How we accomplish that is the challenge.

Keep in touch and PM/email me if you feel like further discussion.

Marci
aka GirlFlyer
  • #16
  • Aug 13, 2007 6:58am Aug 13, 2007 6:58am
It is more than a workable concept...it is a reality. Simplistically, all you need is positive expectancy, reasonable capitalisation, some time and help with the programming/database and then, after project completion, the discipline to not touch and wait.

Sure - Boring after the setup but it works.
  • #17
  • Aug 13, 2007 7:11am Aug 13, 2007 7:11am
Quoting Jason
I trade using a fully automatic system. It actually trades four currency pairs using two different strategies via a broker API. The 'system' runs independantly on a server.

I am certainly no great heroic trader nor a fantastic programmer but I certainly think it is worth setting things up properly to run completely mechanically. The biggest problem for those that like trading and the adrenalin rush that they might get is that this type of trading is completely boring. I do not even have to look at anything unless I receive an email from the 'system' saying something is wrong and that is rare. I then check out the end of month result at the end of each month.

I support a number of points made on this thread. I always find it strange though that discussions like this are generally amongst the most unread. Must be the boredom thing. I think you guys here are on the right track though.
Jason,

I also found it strange that these threads are so unpopular. I think it is skepticism about the feasibility and/or the general difficulty getting them to work. It is a LOT of hard work!

So actually you are the expert on this thread, since you are running live. What kind of results are you getting?

Why don't you autotrade 4 pairs, and then specialize in 1 more for manual trading? Then you can fulfill any manual trading "excitement" that you need, and keep the profits coming from the existing pairs. Personally, I want to be successful autotrading so I can pursue other things in life instead of trading 8 hours per day (or 20 ).

Flyer is right about one thing: this task is very secretive with the big boys. Don't worry though, I haven't spilled the beans on how I build my systems... yet

-Duke
  • #18
  • Aug 13, 2007 12:53pm Aug 13, 2007 12:53pm
Quoting duketoga
Hi jjk2!

Well, it's impossible to say exactly what is going on without seeing the thesis, but it is farly easy to fit the data to pick tops and bottoms. This is especially true if you ignore the timescale, so that each top and bottom is +1000 pips apart, and that maximum downdraw might be -600 pips. In reality, you could never survive such swings, even with enormous profits.

Any system is valid if and only if it works during forward testing. It is very common that a backtested works well, only to fail during forward testing because 1) the model fit to previous data instead of finding generalized conditions, and therefore fails or 2) backtesting accidentally looked 1 period into the future, which skews results badly.

You are right about the complexity of the markets and the unpredictability of trader mentality. But remember also that price is king. All the ebbs and flows of the market, the "animal spirits" if you will, are reflected in the price and the volume! So this data gives us all the information we need to create a model that will trade these sentiments.

Do we need to be 100% accurate? No.
Do we need to know why things are moving as they do? Debatable, but my view is that no, we do not.

We do need to be able to trade with greater than 52% accuracy (or less with higher average gain than loss). The proof is in the pudding with these systems - they are valid when they work!

Take care,
-Duke
the max draw down was 6% where the stops were at. also this is trading future contracts, mostly crude oil. the more liquid and full of speculators, the system adjusts better. for example, the t-bills futures backtest resulted in 98% accuracy.

what basically the thesis is about is taking the regular joe osciallators and indcators, and attempting to correct their "faults." namely the lags are removed, and basically the new hybrid indicator gives a faster readout on the market's sentiment, what the participants are doing. it just tells this with much faster than the regular indicators. of course it will not predict future, sometimes it would pick a top, only to see it continue to go higher. this can be limited possibly by a stop loss. this fake out may or may not result in loss or profit.

i wil post up the forward testing results sometime. but ur absolutely right, backtesting results can be deceiving!

i guess just bad stuff i read about cold war era rocket scientists trying to predict the market. im sure if u got a good model and succeed in making profit, none of the argument matters! your making money! who cares!

but how can i make money with 52% success rate? do u mean by money management ? i guess i could see it work. but it would really be hard to sit through long periods of loss....which defeats the purpose of a mechanical auto system.
{Promotion Removed}
  • #19
  • Aug 13, 2007 1:01pm Aug 13, 2007 1:01pm
Quoting duketoga
Jason,

I also found it strange that these threads are so unpopular. I think it is skepticism about the feasibility and/or the general difficulty getting them to work. It is a LOT of hard work!

So actually you are the expert on this thread, since you are running live. What kind of results are you getting?

Why don't you autotrade 4 pairs, and then specialize in 1 more for manual trading? Then you can fulfill any manual trading "excitement" that you need, and keep the profits coming from the existing pairs. Personally, I want to be successful autotrading so I can pursue other things in life instead of trading 8 hours per day (or 20 ).

Flyer is right about one thing: this task is very secretive with the big boys. Don't worry though, I haven't spilled the beans on how I build my systems... yet

-Duke
no one reads this because this forums not really about autotraders, or whatever. maybe skeptics, but does it really matter? if we have a gem that trades for us while we sit on the crapper while other guys "adrenaline-junkies" are in and out of the market angry because of some central bank intervention, or the markets really crapping out because of subprime? I smile everytime, because I'm not emotionally constrained by such news. As long as markets moving im making money. Only a handful of traders are able to keep their emotions control, so most of us, even with bigger experience are suceptible to emotional bouts, although the ones iwth experience maybe a LOT better at controlling it. regardless, we are humans, we are emotional beings. FEAR for certain, is a big reality.

but overall im really glad i found at least a few others so passionate about thsi subject. could i get u guys MSN or email we can chat more realtime its faster
{Promotion Removed}
  • #20
  • Aug 13, 2007 1:09pm Aug 13, 2007 1:09pm
also on the subject of intuition. i believe you can't prgoram a complete emulation of some of the human traits such as "gut feelings" or "risk taking".

however, if you analyze it carefully enough, its possible to realize and breakdown the subconscious steps being taken. for instance, a month ago, i would never be able to program my system, because it had that "gut-feeling" factor to determine my exit. after looking carefully, i was merely making "gut-feelings" because of arbitrary past experience and prices. then i start make the system to start looking at the past, pick out which ones that gave me the "gut-feeling" to get out, and so on. my point is, such "gut-feeling" intution maybe more logically bound then it actually is. its just a matter of variables and their quantity. of course, i dont have the skills to be using it all, but i think, there are for you computer science folks, ways to solve these type of problems.

of course, this is a mere example, i dont argue that i can emulate a program that can hit on women in a bar. too many variables..
{Promotion Removed}
0 traders viewing now