Introduction & AutoTrading Ideas | Page 2 | 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

  • #21
  • Aug 13, 2007 2:29pm Aug 13, 2007 2:29pm
I don't think I am an expert but I have worked in this way for a while and have spoken to a couple of people I would call experts over the past three years.

My currency pairs are structured to provide a balanced exposure to USD,GBP,JPY,EUR. They all have an equal weight on one 'system' while my second 'system' focuses on GBP/USD only. So I guess all up the GBP and USD are slightly favoured. As for manual trading - sorry, unable to handle drawdowns/emotions and follow all rules/enter all trades without staying up 24hrs and maybe freaking out when I get a bad drawdown run (4 consecutive negative months once!)

My monthly performance consistant with backtested results and really of no importance as it depends on individual risk tolerance anyway (you could double your risk and get double the return and drawdown). The system generates positive return over the long term and automating it makes sure that I don't interfere with my pesky emotions.

It is still interesting to keep an eye on the world economic events however but I don't ever trade based on them (see above personal issues). Not to say that all the discretion based/news guys have got it wrong - just each to their own.
  • #22
  • Aug 14, 2007 6:13pm Aug 14, 2007 6:13pm
Regarding system building and back testing this post by Merlin is an excellent read. Also I found this article ("On Developing a Financial Prediction System:
Pitfalls and Possibilities") which I read a while ago. It touches on some of the
steps involved in building a system and some common mistakes often done, but seen from a machine learning perspective.

The worst and most common problem with back testing is probably over fitting. It is very easy to make a system which is optimized to curve fit the data it is back tested on and therefore fails miserably in forward testing. Idealy a system should be robust to noise introduced to the parameters of the system.
  • #23
  • Aug 14, 2007 6:16pm Aug 14, 2007 6:16pm
I forgot to upload the article...
Attached File(s)
File Type: pdf neural20.pdf   231 KB | 991 downloads
  • #24
  • Aug 14, 2007 6:19pm Aug 14, 2007 6:19pm
Damn it! I also forgot the link to Merlins article. Here it is: http://www.forexfactory.com/showthread.php?t=4446

I am definitely going to bed now.
  • #25
  • Aug 14, 2007 9:50pm Aug 14, 2007 9:50pm
Interesting topic, I just learnt MQL (basically c++, which I don't know lol) a couple days ago.

Anyhow, my thoughts on the subject:

  1. A lot of people may not like it, because it is a newer concept, a change if you will, and you know how the mass population responds to most changes.
  2. I believe autotrading has many benefits contrary to what other people think. It removes all emotions, it will take EVERY trade it is programmed to. All too often a human may or may not take a trade based on an aspect that is NOT part of the system, and that is understandable.
  3. With a positive expectancy, basically an expectancy over the long term for each trade to average out to a positive return, an automated system is like a casino. Casinos have an edge. It may be small, but played out over the course of a year, a 52/48 win/loss ratio (set in stone by the rules of their games) can ultimately result in big gains. I like this article: http://www.investopedia.com/articles...rofit_loss.asp this theory works better than the risk:reward ratio, as a system with 75% win expectancy can have bigger average losses than wins, which is not really accepted by most people in the trading community.
  4. There are patterns on EVERY pair, and EVERY timeframe, though I find the longer the timeframe (daily, weekly) the clearer and more likely these patterns are to occur. Being able to identify these patterns, one is also able to create a program that can identify these patterns. By pattern I simply mean a specific way price moves. A lot of these patterns have well over a 50% chance of completing properly, and, like my previous point, a small advantage (edge) can work out to your advantage in the end, IF followed EXACTLY.
  5. An automated system can run all the time, and is able to take ALL trades that meet your criteria. This ties in with my previous points, a system with a positive expectancy is more likely to play out profitably in the long run if EVERY trade is taken, not select ones limited by "gut feeling", or simply the trader not being available to take that trade, due to work, sleep, etc.

Not to ramble on anymore...

  • #26
  • Aug 15, 2007 3:00pm Aug 15, 2007 3:00pm
Quoting ScottH
Interesting topic, I just learnt MQL (basically c++, which I don't know lol) a couple days ago.

Anyhow, my thoughts on the subject:

  1. A lot of people may not like it, because it is a newer concept, a change if you will, and you know how the mass population responds to most changes.
  2. I believe autotrading has many benefits contrary to what other people think. It removes all emotions, it will take EVERY trade it is programmed to. All too often a human may or may not take a trade based on an aspect that is NOT part of the system, and that is understandable.
  3. With a positive expectancy, basically an expectancy over the long term for each trade to average out to a positive return, an automated system is like a casino. Casinos have an edge. It may be small, but played out over the course of a year, a 52/48 win/loss ratio (set in stone by the rules of their games) can ultimately result in big gains. I like this article: http://www.investopedia.com/articles...rofit_loss.asp this theory works better than the risk:reward ratio, as a system with 75% win expectancy can have bigger average losses than wins, which is not really accepted by most people in the trading community.
  4. There are patterns on EVERY pair, and EVERY timeframe, though I find the longer the timeframe (daily, weekly) the clearer and more likely these patterns are to occur. Being able to identify these patterns, one is also able to create a program that can identify these patterns. By pattern I simply mean a specific way price moves. A lot of these patterns have well over a 50% chance of completing properly, and, like my previous point, a small advantage (edge) can work out to your advantage in the end, IF followed EXACTLY.
  5. An automated system can run all the time, and is able to take ALL trades that meet your criteria. This ties in with my previous points, a system with a positive expectancy is more likely to play out profitably in the long run if EVERY trade is taken, not select ones limited by "gut feeling", or simply the trader not being available to take that trade, due to work, sleep, etc.

Not to ramble on anymore...

you've nailed it. nice job!
{Promotion Removed}
  • #27
  • Aug 21, 2007 5:35am Aug 21, 2007 5:35am
Can a relatively easy hybrid system be created out of technical indicator and AI technique.

I am proposing to use the Fractal up & Fractal down for opening long and short position and uses AI to exit it in the most profitable manner.
  • #28
  • Aug 22, 2007 9:02am Aug 22, 2007 9:02am
ScottH, Excellent points.

We say a 52% win ratio can gain money assuming that each trade has an equal gain or loss. Without the spread, a 50% win ratio will neither lose nor gain money over time, and with a 40 pip average gain/loss (and 3 pip spread), 52% correct is profitable over time. If your average gain/loss is 80 pips, you need only 51% correct.

Of course, as many threads have pointed out, it is also possible to have a lower win ratio, if these wins follow large trends and gain more pips than are lost during loses. In fact, my systems tend to work this way as well.

Are there any others out there with interest or experience in creating automated trading systems? It would be good to get some more insights and experiences on this topic!

Cheers,
-Duke
  • #29
  • Last Post: Edited 6:00pm Jan 6, 2008 8:01am | Edited 6:00pm
I'm very interested in creating an automated trading system myself. I don't know C++ infortunately and I feel like I'm drawing ever closer to the day of learning it but I do know how to code stuff in MetaEditor (pretty much C++ anyway so I'm half way there... or at least I like to think I am) and that's what I'm using at the moment. I'm also a believer of writing the code from scratch and have written one EA in particular that looks very pretty ... mostly kb wise. Unfortunately it needs work to become profitable. It's got some periods where its got blistering performance and then it dies away. But those positive runs are all I need to get me working my arse off to get a working profitable EA. Admittingly though at the moment I'm mostly reading about the different ways of making these systems and from my experience, I can say that its bloody hard (not stating the obvious or anything ) (and hence interesting and challenging ). But I also believe that its possible.

Curious to know what the progress is with your systems guys? Are you starting to see positive results?

I'm also curious to see as to how you guys manage your CPU's? Or are you mainly testing using smaller sized input vectors?

P.S. query

1. The idea
2. The quantitative analysis
3. The control system

what did you mean by that? The idea I assume is the strategy? So for example MA crosses simplistically speaking
Points 2 and 3 I wouldn't have a clue what you meant by them.

The other question which has been on my mind for quite some time now is how to represent n-dimensional data on a 2d or preferably a 3d graph?
0 traders viewing now