Is 100% Mechanical Trading Possible?

[devster 10]

Hi all.. I've lurked a bit, but this comment brought me out of the woodwork..

I'm a 100% mech trader..

 

If you want a real platform for forex there is simply no alternative to Metatrader. Tradestation simply does not compare.

 

Real platform? Not sure what you're doing with metattrader that TS can't do, but then I don't use metatrader..

For me it's pretty simple -it's about practicality - a case of using whatever makes profit. I use TS (ts2000i) and swear by it - it's an autotrading wet dream as far as I'm concerned - I can't see any reason to even look at Metatrader. But I imagine it has the same capabilites, mostly. I have no reason to love or hate metatrader, or any other software - although I hate the TS8 subscription model.

 

Quite frankly I see people getting hung up on their favourite this, that and other.. markets.. software.. brokers.. indicators..

What IS important, is that anyone using software to implement rule based trading stands a much better chance - be it metatrader, TS, wealthlab, even some kind of excel macro - whatever.

The software is simply the tool - it's up to the craftsman to produce the goods!

[Kiwi 16]

I've just started moving towards intraday mechanical trading. I used to mechanically trade futures EOD but stopped some years back to take advantage of the much higher trade frequency in day trading.

 

Money made = expectancy * amount risked * trades per year after all.

 

But I have become bored with the process of semi-discretionary trading. So I am willing to sacrifice some quality for automation.

 

The system will use Sierra Chart for I/O and decision making with Interactive Brokers TWS as the broker interface for trading futures and forex. It will be fully automated.

 

Currently I am developing 3 systems on HSI and will then attempt to adapt them to other indexes and forex pairs. Some may not work on everything. One is countertrend; one is breakout; one is catching the rolling trends. My goal is to be a little fussy about which trades will be taken ... as the computer is not impatient about near misses.

 

Currently it looks very promising but I would say these things:

- no indicators are involved except a couple of mas used when defining trend although I just might try to develop a keltner based system as well.

- because its price pattern based they are much more complex than your average system designers indicator related strategy

- because they are complex there are a lot of variables (system designers are going "warning warning")

-- but the underlying principles are those I've used for discretionary trading for years and the principles are simple ... just complex to program

-- to my pleasant surprise the optimization has proved very robust (a wide range of stops, targets and trailing strategies have proved profitable and surfaces are quite flat).

 

One thing that has kept surprising me is that things I thought when discretionary trading proved to be untrue under testing (once my software got complex enough to do the testing (1000s of lines of code)). It seems that I applied intuition to correct my rules so in some cases things are not mechanizable and in some cases a new rule can be developed.

[forrestang 10]

It is possible. But some discretion has to be applied, and that discretion can too be mechanical.

 

I dont think a system can just be left to run 24 hours a day though and maintain profitability?

[Kiwi 16]

It is possible. But some discretion has to be applied, and that discretion can too be mechanical.

 

I dont think a system can just be left to run 24 hours a day though and maintain profitability?

 

 

I bought a system called GSX in the early part of the decade. It was profitable then with zero interference. And it was still profitable last year with no interference.

 

There is a newer slightly better version but the old one hasn't died.

[Charlton 10]

It is possible. But some discretion has to be applied, and that discretion can too be mechanical.

 

I dont think a system can just be left to run 24 hours a day though and maintain profitability?

I think this is exactly right. The development of a mechanical trading system is an iterative process that involves both discretionary and mechanical elements.

 

You start with a concept or strategy that you have traded in pratice. In order to mechanise it there is a need to identify a set of rules/logic that can be converted into a programmeable language. This process will also help clarify the manual system (the starting point) thereby helping it to become more successful as a by-product of the exercise. Equally it will identify aspects of the manual system that are currently discretionary.

 

Taking a simple example, you might have a trend following mechanical system that is fully mechanised. However it might only work in a trending market, therefore there is a discretionary element that needs to "switch" it on and off depending upon market conditions. In time, using for example different indicators, you might be able to identify in a mechanical way whether the market is trending or consolidating, in which case you have created a mechanical switch that can sit on top of the trending system.

 

The mechanical system can then be further developed to include components for placing trades during consolidation periods.

 

Theoretically I see no reason why a mechanical system cannot be left on 24 hours and maintain profitability, but this is subject to the switching concept I mentioned earlier and the development of trading components that handle each market characteristic. This would not be easy in practice, but it is entirely logical.

 

However in practice, given the limitations that retail traders face, we probably have to settle with a mix of discretionary and mechanical trading coupled with a constant feedback between the two, such that both types of system improve over time.

 

Charlton

Edited September 8, 2009 by Charlton

[ZOSO 10]

A 100% mechanical system is not only doable, but should be on every trader's to do list.

 

A mechanical system with a positive expectancy is the closest thing you'll find to the Holy Grail. If it is optimized to run 24/7, all the better. It's like owning a casino without having to bring the suckers through the door.

 

Just make sure you are adequately capitalized for those inevitable draw downs.

Edited September 8, 2009 by ZOSO

[maxima 10]

-- but the underlying principles are those I've used for discretionary trading for years and the principles are simple ... just complex to program

Finally we can talk not having angry Russians on the back

 

Kiwi your underlying principles are simple - for you - because you've been polishing them for 10 yrs. On other side - I guarantee (I am expert software developer with 20+ yrs of experience including City) - that 'complex to program' is because you stuck with Sierra. Programming Sierra is same as running 40 km maraphon in Dr Marten boots and prickly Jesus hat with sandpaper tightly bound to your crotch.

 

My point is - if you tried other platforms - you might find your programming experience crucially changed, bringing you much easier, controllable, faster and reliable result.

 

I understand you bored with your success. Lets pepper it a bit but if you'd like the best program you can - consider other posibilities.

 

Hope you understand I tried to help and not having a rant on SC. they are dead anyway. Beware complacency!

 

:missy:

[forrestang 10]

Finally we can talk not having angry Russians on the back

 

Kiwi your underlying principles are simple - for you - because you've been polishing them for 10 yrs. On other side - I guarantee (I am expert software developer with 20+ yrs of experience including City) - that 'complex to program' is because you stuck with Sierra. Programming Sierra is same as running 40 km maraphon in Dr Marten boots and prickly Jesus hat with sandpaper tightly bound to your crotch.

 

My point is - if you tried other platforms - you might find your programming experience crucially changed, bringing you much easier, controllable, faster and reliable result.

 

I understand you bored with your success. Lets pepper it a bit but if you'd like the best program you can - consider other posibilities.

 

Hope you understand I tried to help and not having a rant on SC. they are dead anyway. Beware complacency!

 

:missy:

 

What would you recommend?

[gooni 10]

NinjaTrader is good for beginners but quite limited. NeoTicker has an excellent framework (especially if you want to do things that involve multiple instruments or timeframes), but can be frustrating to learn. Or go straight to the API.

[Kiwi 16]

I don't think the issue is so much Sierra Chart as what I'm doing. My model of the market is one that is easy to see by trained eye but difficult to clarify to the extent that I would like. However I now have two of my three models looking reasonable.

 

One has a pf of 2 and a win rate under 50% on out of sample data. The second is over 3 with a 60% win rate out of sample. The third, which I left to last while I worked through the issues with the less promising options should hopefully eclipse the other two.

 

I know SC rather well and doubt very much that I'd find any other platform easier (I did program a few earlier and periodically discuss strengths and weaknesses with friends who use them).

 

On the issue of running a system 24x7 I wouldn't choose that - but not because you couldn't, more because, in my experience, the markets I am interested in all have times when their behaviour favours one strategy or another. For example, what trend trader would day trade 6E during the US morning if they had the choice of the less news shock prone European morning instead.

[maxima 10]

well, I tried

 

anyway I was curious last night (I dropped an automated system idea a year ago so I didnt do much of programming lately) and did simple trading system using your special MAs you published on Sc recently... I spent 30 min to programm it using Sc website and then an hour trying to backtest with the replay feature...

 

Single threaded aplication in 21st century! gve me a break!I couldnt get even one single result I accept as reliable..

 

I have no idea what patience you have to test a real money system.. Anyway. I havent seen much of platforms (Xtrader and CQG (excellent platforms but dont know the programming capabilities - APIs are too expensive - never tried), TWS - not sure they have).

 

However I used Ninja. their analyzer could have produced few hundreds of reliable result sets, but I am biased here as C # is my field of expertise. Unfortunately Nt doesnt have static DOM so I stuck with Transact and Sc for now...

 

The comment above that Nt being for novice is BS because you can connect your own .Net (or even C++ app) to the NT and use whole power of .net or whatever. I know cos i did it year ago. But for manual trading it is an average system, perhaps new version will be better. Xtrader is the very best one. Way better than self made systems in the UK banks.

Edited September 12, 2009 by maxima

[Blaiserboy 10]

Someone is a little touchy. It was simply a comment based on my trials and tribulations. If you want a real platform for forex there is simply no alternative to Metatrader. Tradestation simply does not compare.

 

I have used Metatrader for a few years and find no great problems with it, although I understand Tradestation and Amibroker to be superior for developing and testing automated systems... I just started with Amibroker for the purpose of learning re automated systems, it seems easy enough to use, I haven't used it enough to comment on the testing. Regarding costs...Metatrader is free, as is Ninja, each has a couple drawbacks... and the others cost and have some unique advantages.

[Kiwi 16]

Maxima, the trick for me is that I had already written the order entry fron't end while I was partially automating my day trading. Then, I started to use Amibroker (which I like) to test systems but ran into two issues:

- I didn't want to use it live (I'd much rather trust SC for that)

- Its hard to do multi-timeframe stuff with it but easy for me with SC

 

So I had the choice of building on Amibroker and moving to SC or building a test harness on SC and doing the full development on SC (admin will have a test harness next year thats better than mine but I can get the key elements of multivariable testing that Amibroker gave me from my version).

 

It seems that everything I look at is developed with some group of traders or some paradigm and none quite fit my internal model so whatever I do I'm going to have to build some of it myself. LOL. So I stick with the one I know really well --- better the devil etc.

[forrestang 10]

Maxima, the trick for me is that I had already written the order entry fron't end while I was partially automating my day trading. Then, I started to use Amibroker (which I like) to test systems but ran into two issues:

- I didn't want to use it live (I'd much rather trust SC for that)

- Its hard to do multi-timeframe stuff with it but easy for me with SC

.

 

Can you expound on amibroker? What if I didn't want to use it to actually RUN an automated strategy, but is it good soley for backtesting, i.e. the ability to give a more thourough backtest than ninja?

 

What language can you write the program in for Ami? As I've written a few things in Ninja, it would be nice if I can port it over w/minor changes.

 

And why is it that you wouldn't want to automate anything on Ami?

[Tams 20]

traders.com has articles with source codes in different languages.

you can easily compare the complexity of one programming language with another.

[Kiwi 16]

Amibroker has a language that has similar structure, syntax etc to C, C++, C#

 

In fact it is close enough to Sierra Chart's dll interface (C, C++) that I was going to test with AB and move to SC. Its also a very fast backtesting platform.

 

But AB is more about single timeframe testing than anything else and I don't trust its interface with Interactive Brokers.

 

By contrast Sierra Chart is very good for multi-timeframe and for communicating information between charts - so it suits the style of what I know I can make work. Also SC has a robust trading interface with IB which I have used for hundreds, maybe thousands of trades so I know how to drive it and trust it with my money.

 

Note: a lot of platforms are a lot like C in their underlying language (Amibroker, Sierra Chart, Ninja (I think) and ESignal; metatrader stuff has a similar structure too). They then add interfaces to their charting and order entry and monitoring. Much of the complexity and difference lies in this area ... but once you learn one then if you can make it do what you want it tends to be easiest to stick with it.

 

Once you can do it then whether it is rich enough to handle what you want becomes a big issue.

 

At the moment Sierra is relatively week on testing but very strong on real time. By mid next year I expect they'll be medium/strong on testing. At present Amibroker is excellent on testing but the real time interface feels very beta and lacks richness. Horses for courses.

 

 

Tam is right about traders.com except that they tend to be doing "yet another indicator" and its very different to building a trading robot that you can trust to run in real time. For that I suspect Tradestation, Metatrader and Sierra Chart might be the more mature choices - but I don't have wide experience of other platforms (and my endorsement of Metatrader is not based on personal experience, just the number of people at FF doing stuff with it).

[forrestang 10]

Amibroker has a language that has similar structure, syntax etc to C, C++, C#

 

In fact it is close enough to Sierra Chart's dll interface (C, C++) that I was going to test with AB and move to SC. Its also a very fast backtesting platform.

 

But AB is more about single timeframe testing than anything else and I don't trust its interface with Interactive Brokers.

 

By contrast Sierra Chart is very good for multi-timeframe and for communicating information between charts - so it suits the style of what I know I can make work. Also SC has a robust trading interface with IB which I have used for hundreds, maybe thousands of trades so I know how to drive it and trust it with my money.

 

Note: a lot of platforms are a lot like C in their underlying language (Amibroker, Sierra Chart, Ninja (I think) and ESignal). They then add interfaces to their charting and order entry and monitoring. Much of the complexity and difference lies in this area ... but once you learn one then if you can make it do what you want it tends to be easiest to stick with it.

 

Once you can do it then whether it is rich enough to handle what you want becomes a big issue.

 

At the moment Sierra is relatively week on testing but very strong on real time. By mid next year I expect they'll be medium/strong on testing. At present Amibroker is excellent on testing but the real time interface feels very beta and lacks richness. Horses for courses.

 

Wait, I have Sierra also, and was looking to do some back testing, but it seemed like Sierra was annoying?

 

Here is what I mean. I was looking at the videos on SC's website to program an automated strategy, and they had examples using that Excel type strategy creater? Now I'm ok with Excel, but writing a strategy with it is something I don't even want to try to do.

 

What was good about SC over ninja though, is that you can backtest the ACTUAL tick data which was great!!!! Ninja requiring the multi-time frame programming to get decent resolution for back testing is cumbersome.

 

Am I wrong about the way to write strats in SC?

[Kiwi 16]

Yes.

 

You use what Anthony calls ACSIL (Advanced Custom Study Interface and Language) which is a rich set of interfaces to a C++ compiler (the watcomm one is delivered built in with the program but some people prefer other compilers/development environments). You can then write whatever level of complexity you require to drive your trading program.

 

Basically there are two ways to code for SC:

- the worksheet approach which is more intuitive for non-programmers but not as rich or flexible

- the ACSIL approach which will let you do anything but is more "advanced."

 

For simpler strategies both will give the same result.

[Charlton 10]

Can you expound on amibroker? What if I didn't want to use it to actually RUN an automated strategy, but is it good soley for backtesting, i.e. the ability to give a more thourough backtest than ninja?

 

What language can you write the program in for Ami? As I've written a few things in Ninja, it would be nice if I can port it over w/minor changes.

 

And why is it that you wouldn't want to automate anything on Ami?

It's very easy to get a feel for the Amibroker offering from their site. Look at the AFL library AFL Reference links on the main web-page - they show the code used. Also there is a link to their Yahoo User Group. In addition you can download it for a free trial (1 month I believe). You could probably test out your NinjaTrader exports in that time. Also worth looking at is the Knowledge Base with complete code examples.

 

I still have it but don't use it, because I use Tradestation now. Like Kiwi I do multi-timeframe analysis. AB does appear to handle multiple timeframe analysis e.g. Multiple Time Frame support, however it depends upon what precisely you are trying to achieve and clearly it did not satisfy Kiwi's particular requirements at the time.

 

There are certain issues in TS relating to multiple TFs, which may require the use of globalvariables or even external functions. Some users will hit these issues because of the particular needs of their strategies, whereas others will happily use a particular platform/proprietary language blissfully unaware of/not effected by them. In truth this is difficult to predict until you start using the software in earnest.

 

Charlton

[maxima 10]

So I stick with the one I know really well --- better the devil etc.

Totally understand...

 

I didnt mean to start the-best-programming-language ideological war in this thread sorry!

 

Multiframe is a good argument. You can have MF in Ninja - you cant chart it though but automatic system can have. NT promised MF a year ago. Last news they said - totally new version including MF is coming this month.

[Tams 20]

MultiCharts is very capable with multiple time frames.

You can even have "time" and "non-time" based resolutions in the same chart.

 

e.g. one data series in 5 min resolution while another data series in constant volume bars.

 

 

However MultiCharts cannot "cross-talk" between charts. You will need a 3rd party add-on to handle the communication (GV, ADE, etc.,) , and the process is complicated.

Edited September 12, 2009 by Tams

[RM686 10]

what if you dont know programing? are there any mechnical systems for charts and ETFS that one can buy.

[Tams 20]

what if you dont know programing? are there any mechnical systems for charts and ETFS that one can buy.

 

 

what is more difficult to learn than to learn to trade ?

if you can learn to trade, you can learn to program in EasyLanguage.

[sevensa 10]

what if you dont know programing? are there any mechnical systems for charts and ETFS that one can buy.

 

Yes, there are 1000's. Just google, or check out e-bay. There are numerous system that promise to make you a millionaire in 6 months.

[RM686 10]

what is more difficult to learn than to learn to trade ?

if you can learn to trade, you can learn to program in EasyLanguage.

 

Reminds me in grade school when you couldn't spell a word the teacher would tell you to look it up in the dictionary. You would never find it if you thought it was like KAT. Some peoples can't spell and will never be able to. Some people can sing or play the violin and some never will even with a gun pressed to the side of their head.

 

I love the other answer that there are tons "on the internet offering to make you a millionaire". Of course thats true and their all frauds. Very helpful thank you.

 

I asked a simple question and I am assuming the answer is if you are not a good trader and can't program you are out of luck. Although none of you directly answered my question in a round about way I guess you did.