MA Cross Optimization EA (very cool)

This is very interesting.

Can you do a test with 2 LWMA crosses with 5 LWMA?
I have always believed in MA, been searching for ways for filter the bad signals.

Thanks

[quote=tdion;2106919]Say whatever you want to.... this is impressive..... Again, I was bored so I ran this thing on the USDJPY for about 60 seconds..../quote]

It will work as a toy to relieve boredom but the only thing it will teach is that the market is dynamic and does not continue to repeat the past. You can keep optimising and adding filters and re optimising until eventually you will have a virtual holy grail had you used it over the test period. the more data you test the more over optimised the result will be. Do you really believe that this will work in the future? dont you think if it were possible to number crunch the perfect trading system it would have been done by now? If this is possible, how come hedge funds and even banks lose, with millions of dollars to spend on cracking market codes? all you will achieve is a curve fit on past data and a nasty lesson from the market if your silly enough to think it can be beaten as easy as that

Hello
I am completely new to all this well FX anyway.......... did trade stocks for a while .. but its not my background......... I have looked at various threads with great interest..... one thing that I really can see is that a number of people seem to think that there is a holy grail systems out there....... I just cannot see it....... as its been stated .. after all this time .. someone or bank would have found it ....

I only look at GBP /USD , EUR /USD JPY/USD CHF/USD

I ontry trade EUR / USD , JPY/ USD

I only use ma cross overs to alert me when the prices are near the 50 day moving average and are in over bought over sold conditions .. also .. I look at the charts and see what the general trend is for the USD......... all facinating stuff..........

hmmmm..... i like to explore facts

here is what is known at this point.... the USDJPY graph is made of 700 trades over 30 years, analyzing daily price movement. MACDoug i agree that it is a curve fit of data.... but step outside the box, man!

we have a set of rules that fits 260X30 days, or 7800 data points. i never claimed to find a holy grail. as a scientist, i claim that the characteristics of this particular market over the last 30 years can be closely mapped to the 16/18 cross.... there is no disputing that.

the bottom line:
the reality is, as long as USDJPY's characteristics continue the same way they have over the last 30 years, the cross will work. if the characteristics change, the cross will fail. do with it what you want to.

I'm wondering, tdion, so you know of any self-optimization function in metatrader?

I noticed the IsOptimization function, and I'm guessing there is some sort of metric for determining whether or not optimization is running...

I figure that if we can get the EA to continuously optimize itself over the past year or so to find an area of probability for the coming month.

I'm personally looking into it, and any other programmers interested are welcome to join in, I'm following these instructions: http://articles.mql4.com/506

i know nothing about self optimization of MQL code, but it wouldn't be hard to do.... if you stored the previous months candlesticks in an array, and reoptimized within your own MQL code...

i think your idea about reoptimizing every month has merit.... and is worth exploring.

heck, you could optimize the optimization period (ie how far back to go) but then i think we'd be getting deeper into curvefitting than we'd want to.... but then again....

however, what is truly impressive about the 16/18 cross is how it stood the test of time with no changing of parameters. hell, this thing would have been making consistent profit for my entire lifetime (i'm 32) and that's a long time.

there was a garbage pale kid named ray gun. one called rapping ron, too... both were reagan knockoffs. maybe you've heard of garbage pale kids.... they were the pokemon cards for my generation.

http://i1.iofferphoto.com/img/113860...10362759/1.jpg

this 16/18 cross by itself is too raw to trade. what i am thinking about doing is taking 1 or 2 other profitable USDJPY systems that are not related to price alone (a neural network of fundamentals where price is the output function) and building a system where when 2 or 3 all show buy, then it's a long. same for short.

indeed, a MA cross system that is always in the market is a dangerous thing. however, it does show some potential to identify the USDJPY market characteristics, which could serve the purpose of acting as the component of an underlying system.

Good point IMHO. What you've identified is that the UJ has "trended" in a certain cyclical pattern, with a decent measure of consistency, over a 30+ year period. MA cross is merely a tool, and in itself is not necessarily a dangerous thing. What is potentially dangerous is the assumption that any kind of past behavior will continue to repeat itself accurately enough to conform to a model that is overly specific. That is not necessarily restricted to indicator-based entry methods.

Dynamic optimization might not necessarily be the answer. There's a product called Best Charts (see http://www.stock-anal.com/) which I trailled extensively some years ago, when I was trading stocks. It's based on optimizing crossovers (profit-wise) of approx 15 different popular technical indicators, and then summarizing the component results (i.e. optimized indicator above its signal line = +1, below signal line = –1) into a master indicator.

One effect one had to live with, was that what was (for example) a buy signal X days ago, would not have been a buy signal now, based on the latest re-optimization. I suspect that this effect was exacerbated by the fact that Best Charts used a limited "rolling" window of data, i.e. earliest prices were being discarded to make room for the latest, on a daily basis. (This limitation was supposedly to allow reasonable response time, given that the product was back-testing maybe 500 combinations of indicator/signal line values x 15 indicators x a portfolio of different stocks).

A closely related question is obsolescence, i.e. to what extent recent behavior is more "representative" (and therefore should be weighted accordingly) than earlier behavior, in terms of the immediate future.

IMHO the success of this approach ultimately depends on how consistently prices move in cycles, and how frequently/non-randomly these cycles change....... anyway, more on this later, gotta rush now..........

David

not only Ed Sykota but a lot of other traders claims have made fortunes using MA crossovers. There is a guy* here in Brazil (one of the biggest gurus here) that says that made more than $10.000.000 using only MA crossover.

in the past when i was playing with Metastock (i don't know how to code but with Metastock is easy to edit parameters and indicators) almost all systems with overoptimization gave almost the exactly same results instead of one wich i will send in pm to Tdion test

* this guy broke at least 3 times

ronald i'll say you really got my mind going

i started finding the optimal H1 setting for each month, and then took the REVERSE POSITIONS of the best cross the following month...

i basically bet that lightning won't strike twice in a row

there's a lot here if you sit down and think about it.... that is, there are many applications for this simple tool.

Lol, glad to be of service.

MA crosses have traditionally been used to determine direction. Can they predict strength?

I went back over a lot of the rainbow scalping techniques, and I think a measure of how many MAs cross in a given period can determine trend/breakout strength. The probability of certain distances being reached before the first retracement can be measured, and lots sized accordingly...

Just my idle thinking.... =P

after running about 20 more optimizations, a new truth became apparent to me (why it didn't before now i don't know)

i have been testing all crosses between 1 and 30 for all pairs on various timeframes.... where there are exactly 450 different combinations (1/2, 1/3, 2/3, etc)

usually about 100 are profitable, with 4 or 5 very profitable. so of all the random walks the market could have taken, a tiny handful would have maximized it.

it reminds me of the lottery.... pick two number between 1 and 30, and you have a 5 in 450 chance of to compound the hell out of it. if you guess right, you'll make a mint. but unfortunately the other 350 unprofitable crosses will always be there to remind you of reality.

this little tool also shows me that creating curve-fit EAs is a piece of cake, which is probably what a lot of Ebay EAs are based on.

Tdion,
Have you optimised different type of MA's,I mean adaptive,Hull,Jurik etc..Or Have you optimised one type of MA with different setting?

so much for confidence in 8 yrs of data
you are chasing your tails but I think Ive made my point. Ill leave you guys to it

Here are Hull, Jurik if you want the formulae.........

David

Thanks

maybe Tdion already know that but i will post because can help others.

those are supossed to be the "correct" T3 and Jurik MAs. I remember that both were posted by Newdigital at ForextTSD.

look at the sharp ratio of this 6/8 USDJPY daily cross from 1979-2008.....

it only made 15,000 pips (compared to 22,000 w/ the 16/18 cross) but i thought it was worth sharing.

Not until the trend has already used most of it, no. The divergence between the 2 MAs is effectively the strength of the trend past. Altho it can be quantified by using the MA it's resting on as a measure.


Nah. I've had really good luck w/ a 3 EMA and the 9 EMA on the 15m intra-day. It doesn't predict momentum or trend strength, but if you restrict trades to times where momentum is most likely... it works out pretty well. Still, I don't use crosses for entries, I use them for exits. A little lag is beneficial to exits, where entries... not so much. A 5/8 is a very common alternative too.

Since you like larger TFs take a look at the 3/9 or the 5/8 on the monthly as both entry and exit.

the 18/16 one more time, with a 975 TP and 300 SL.... (thus, not always in the market 100% of the time)

profit increased 3,000 pips (to 25,000) w/ the TP and SL....

maximum drawdown only 1,700 pips