Why do most EA's fail ?

[soundfx]

Hi Guys,

 

A recent comment from Rio got me thinking in the thread:

 

http://www.indo-investasi.com/viewtopic.php?f=59&t=8075

 

Where he made the profound statement:

 

Don't get too excited. Most EAs are crap.... remember that.

There's a reason for it.

 

I didn't want to clutter that rapidly growing thread with discussions on why this should be so, hence the new thread.

 

One of the main rules to be drummed into manual traders starting out is always "Be disciplined!".

 

So...what can be more disciplined than a computer program ?

 

Doesn't that mean that any automated system should outperform virtually all manual traders because it has 100% discipline in blindly following the rules of a good system (i.e. one with a consistent edge) ?

 

If we toss a coin and call "heads" we have 50% chance of this being correct in the long run according to mathematics. In theory we should be able to enter the market and call "Long" at any time and have a 50% chance of being correct in the long run.

 

If our trading system has targets (reward) which are twice as much as out stops (risk), then we will always win in the long run.

 

Why are most of the robots out there so rubbish (despite the marketing hype lol) ?

 

What do we need to do to make a robot which works (most of us have heard of great secret robots which are traded by a select few traders which make consistent profits) ?

 

I have a few ideas of my own as to why this is not as simple as it first seems, though I'll be interested to hear comments from you guys. Maybe with the power of many brains we can crack the secret code to allow us to create a robot which will be consistently profitable.

[Cyrillic]

Re: Why do most EA's fail ?

 

Ea's are crap because a good ea takes alot of optimization and code to make it good, or a nice idea to make it profitable still. Why do 95% of manual traders fail? Because 95% of all people fail, people come into this market expecting to win 100000 and retire to the islands, few want to actually make it their life work. People do not wish to spend their time learning this market they want "Gurus" to give them everything. There are very few Gurus if any that will share their experience with you and even if they do people would rather listen to a poser with lure of quick riches than to a person that knows a few things and actually makes money trading. People like to be a in crowd even if crowd is a bunch of long term losers (example most of the forum up online are filled with people that say "Well you have alot of experience why don't you share it with us" and if you do they tell you "well i know this and this and what you are telling is wrong and won't work in a long run" even thou they were the once that asked for advise first. Same with ea's why not share 100's of hours of hard work with "kind" people of forex. Because your ea will get bashed and destroyed since people online have looser mentality and do not want you as a trader to succeed, you do not have 50% chance to win by default in this market because of the spread, which means if you toss a coin eventually spread will eat your deposit. It is also funny that people say you have 50% chance to be right, but 95% of people fail. It means that statically you have 5% chance to be right (funny i know) I believe in this forum and sharing things on it because its a place where we share fraud to determine one and buy good products which we test first. This forum has tons of good ea's and most of the people here do not use it accordingly, people that say (most ea's are crap) are right, but if people start looking through this forum, find and test every ea they find, they are bound to find ea's that are actually good. Hope this has been helpful to answer your long post.

[soundfx]

Re: Why do most EA's fail ?

 

Hi Cyrillic,

 

As a respected trader and poster I appreciate your thoughts ^:)^

 

Good point about the spread (which is so easy to forget in these sort of theoretical thought experiments). However, spread would pale into relative insignificance if we have a very good risk/reward ratio.

 

I agree that the problem that we have in II is that we have loads of members searching around the web for the next greatest thing (normally another over-hyped EA), and eventually the EA appears and often the thread goes quiet - no backtesting or forward testing results from the original poster, so it's often hard to tell which EA's are good.

 

Looking for threads with the most posts can be a good pointer to a reasonable EA, though it's still very tricky to sort out the wood from the trees and find the better performing EAs. Also post count may mean nothing, it could be just lots of folks complaining that the EA doesn't work as expected ;)

 

You mention a lot of optimization being required to create good EAs - though in my book this is "curve-fitting" isn't it ? A good EA should be able to adapt to market conditions whatever they are (we can do this by using ADR, S/R levels etc. rather than fixed stops and targets as a starting point).

 

Your answer was very wide ranging, I was expecting answers to be more along on the lines of answers like:

 

"EAs are typically designed to trade trending markets and fall apart in ranging markets"

"Most EAs using lagging indicators, so will always fail in the long run" etc.

 

It seems to me that an EA which trades one a day should be relatively easy to make - but what's the point? lol, we may as well just set Buy-Stop and Sell-Stop orders at prime breakout levels.

 

EAs as I see it only really come into their own for very short term trading, however I've yet to see a decent scalping EA. All scalping EA's I've seen have target 10-20 pips and a whopping great stop of 200-600 pips =))

 

I think that the key difference in scalping is that when manually scalping you sense very quickly when a trade is going bad and get out quick, well before the stop is hit. However an EA is most likely to be set up based on manual trading parameters which could be 10 pips target and 20 pips stop. We need a very high accuracy on our entry signals to even stand a chance of staying in the game, because the EA will always take us out of bad trades at 20 pips (or worse from slippage).

 

I admit to being a skeptic on occasion myself after being bombarded with so much marketing hype, though I agree that the loser mentality of some on-line "trolls" can hinder the progress of the rest of us (the good news is that other forums are much worse than II). There's enough cash in the Forex markets for everyone on the forum to be hugely successful. The more successful traders who are attracted to the forum, the more the success will ingrained in everyone else :)

[litxus82]

Re: Why do most EA's fail ?

 

Cyrilic,

 

Can you tell us what do you think are good EAs in your opinion?

 

soundfx,

 

I do believe that most EAs which are not like "hacked" type will fail sooner or later as parameters stay the same but markets change. To keep inline for regular EAs (non martingale like scalpers and others) frequent optimization would be needed, and for martingale type EAs they will always work you just need the balance to handle worst case scenarios and best to start with smallest lots possible (nano lots).

[Rio]

Re: Why do most EA's fail ?

 

Most EAs are rubbish because they are written by programmers with very little understanding of what drives the market. Secondly, a robot cannot use fuzzy discretionary logic. Thirdly, the more complicated the system, the harder the testing and optimization... therefore some manual indicaor-based EA systems simply cannot be backtsted for reliability.

 

It seems to me that the real millionaire traders are the ones who simply use fibs and chart patterns,with a good measure of understanding of support and resistance in order to trade. No other indicators are really necessary.

 

Also... in order for accurate entry and TPs, you need fibs. No questions asked. You can't run with systems that have a set number for take profits and stop loss. It's crazy. Pro traders KNOW where the price is going to stop and turn around from technical analysis (fibs), and it's at these levels that TP and SL are set.... NOT some optimized order set up, because market price ranges change and only fibs can track them, whereas optimized values will, one day, cause the EA to go bad.

 

With a good enough entrand the right moment, you can have VERY low risk trades and have very small stoplosses,... because you KNOW that the price will not go beyond a certain value. Most EAs simply don't take this into account.... because they are based on indicators... which are discretionary and innaccurate.

[soundfx]

Re: Why do most EA's fail ?

 

litxus82,

 

Thanks for your comments.

 

I'm not a big fan of the martingale/grid/stacking EA's or whatever they call themselves, though I agree that if you like a roller coaster ride from running an EA you have to keep the stakes very low. Those EAs can make a stack of cash one month and blow your account the next if you're not extremely careful.

 

Rio,

 

Thanks for joining in the discussion as your comment started this ;)

 

You've raised some good points, I'm not aware of the trading calibre of most EA programmers, though I suspect that the best traders just trade and have no interest in programming.

 

Regarding the fuzzy logic. I've often wondered if there's a way to program some of the "gut feel" of trading into programs. Trader's "gut feel" isn't normally just an emotional reaction, it's more like a second sense based on seeing many similar setups in the past. Some traders will say "I didn't like the look of the price action, so didn't take that trade with the trend" for example. If questioned more deeply we could find that this is based on two thoughts:

 

1. The earlier trend has been evening out, so this could be the end of the trend.

2. The spike candle is looking like a reversal is near.

 

Through some research, we should be able to assign reasonably accurate probabilities to each of these events.

 

P(both events) = P(Event 1) * P(Event 2)

 

In other words:

 

P(reversal) = P(trend evening out) * P(opposite spike to trend)

 

Which could be:

 

3/4 * 7/9 = 21/36 = 58% chance that the trend is coming to an end.

 

Perhaps we could use probabilities like this in sophisticated EAs - just a thought...

 

Unfortunately, that leads into your third point which means that backtesting and optimisation will become much trickier.

 

It seems to me that the real millionaire traders are the ones who simply use fibs and chart patterns,with a good measure of understanding of support and resistance in order to trade. No other indicators are really necessary.

 

I agree with that comment entirely. Interestingly, I don't think I've seen an EA yet which looks for specific chart patterns like pennants, flags etc. A pennant for example is a point of compression and in theory it would be simple to see if a we have a pennant and then set buy and sell orders to pick up the breakout which is sure to follow (though we're not sure in which direction, hence the orders).

 

I agree also that using S/R levels and Fibs is a great way to optimise our entries to a high degree so that stops can be kept tight. Keeping stops low, with a good edge on trade entries would make a good EA in any timeframe.

[fork4k]

Re: Why do most EA's fail ?

 

Very interesting discussion!

 

I've often wondered if there's a way to program some of the "gut feel" of trading into programs.

 

Yes, I agree with you there. I believe "gut feel" can indeed be programmed into an EA by really drilling down and asking the right questions in order to translate the gut feeling into the objective, but unconscious observations made by the trader to come up with his or her gut feel. However, I would think that this is not as simple as that because the traders gut feel more than likely evolves without him or her realizing. So while today his gut feel arises from evening out and spikes, a few months later it might arise from a different reason. And it'd be hard to find that out without continuously questioning the trader.

 

Regarding the comment on millionaire traders using fib and chart patterns, I totally absolutely agree too!!! When I read that comment I was like "BAM ... hit it on the nail". And soundfx, the reason why I believe we don't have many (any?) EA's that reliably trade based on chart patterns is because programmatic pattern recognition is HARD. Only today are we starting to see some useful image recognition software coming out on the market (eg: ones that can reliably detect and crop out faces without human intervention or suggestion to the software). Will pattern / image recognition always be hard? I don't know. But, I do think it will be very costly to implement, at least in the foreseeable future. Creating an EA based on indicators isn't terribly hard when you compare creating an EA based on pattern recognition ... at least recognition to be degree that a human brain can recognize a pattern. And if someone were to invest the time and money to create such a pattern-based EA once the technology of pattern recognition becomes reliable ... will it be profitable for them to sell the EA? I doubt it ... we'll have to see.

 

So, I guess my contribution to the discussion is this:

An EA without oversight and manual intervention probably wouldn't work for long because it is technically expensive, if not impossible, to create one close enough to the pattern recognition and adaptability of the human brain and unconscious mind.

[soundfx]

Re: Why do most EA's fail ?

 

Hi fork4k,

 

Thanks for your contribution and some more interesting comments.

 

You've raised a good point about "gut feel" potentially changing over time, however there should be some basic checks which we can make in code which will enhance the performance in general - I strongly believe that spikes are a very powerful price action indicator for example. However, I agree that there's no way we can ever get code to trade any where near as good as a competent manual trader, though we may be able get a bit closer than we currently are.

 

I can see that creating an EA with lots of complexity will be expensive in time and effort, though again rather than try and do everything, we could start with some simple pattern recognition algorithms.

 

Head and Shoulders is a pattern which is simple to see, though would be very tricky to implement, however pennants seem to be quite straightforward (in theory).

 

For a pennant we should check that the preceding price action before the compression was directional i.e. reasonably strongly Long or Short (and not ranging). To check the compression, we look at the recent High-Low average bar range and compare that to the previous High-Low average bar range. If this is much lower then compression is occuring.

 

To identify the pennant formation itself we just look at the last say 3 bars. We can take a linear approximation of the points on all the highs and similarly for the lows and if the high line is pointing down and the low line is pointing up then we have a pennant.

[Rio]

Re: Why do most EA's fail ?

 

The only EA that I know of that does chart pattern recognition is the winner of the Forex World Cup 2009... the LMD multicurrency EA.

 

I don't think it's impossible to code such an EA, but very difficult. There's a fair bit of fuzzy logic in there to do pattern recognition.

Best place to learn about patterns, I think, would be in the James16 group (google it), and all about fibs from the Bobokus and Skunny threads on Forex Factory.

 

In any case, indicator based EAs (because most indicators, especially the averaging ones, are discretionary), and EAs that use set pip take profits (say a set take profit of 30, 40 pips) are doomed to fail when market ranges change.

 

Then we have all the EAs that have small take profits and huge stop losses (a big NO NO!), and the martingale EAs (which blow your account up), and we can really whittle down the list of good EAs.

 

Price action is king, and when price starts exploring areas that it's not supposed to, we get solid trade signals.

[sheetal]

Re: Why do most EA's fail ?

 

Hello friends,

 

I don't have the amount of experience in trading like you guys here and am just a month old kid in this business. Well, in my finite capability of understanding things and the markets :-?? , what I understand is:

 

Markets are not definite and it's not like a problem of algebra in Mathematics that could be solved by implementing a particular formula/formulas. It's not precise that when the London breakout happens, the price would shoot precise 100 pips each n everyday. It's variable / random. And this is nothing but the result of human emotions.

Now EA(s) are nothing but computer programs. And to my (finite) knowledge, computer programs or even the most sophisticated computer systems cannot measure/interpret human emotions. It is a fact that markets are driven by human emotions. Now a gap occurs when a machine or bunch of machine codes tries to interpret human emotions - it's way tooo complicated :-/ for them to understand whereas a living, breathing normal human being could easily understand/interpret another human being's (or a group of homo sapien's) emotions.

 

That's all. Let me know if this makes any sense to you guys

 

cheers :-bd

[soundfx]

Re: Why do most EA's fail ?

 

Rio,

 

Thanks - I'll check out LMD and those FF links.

 

You've nicely summarised how to filter out the dross EA's too in the search for the best.

 

Sheetal,

 

Thanks for your comments.

 

I agree that markets are driven by a force of combined human emotion, however that "group emotion" isn't something which is totally random and impossible to predict (i.e. the market isn't a "random walk" as some academics amazingly still believe!), otherwise none of us would be trading at all :)

 

You gave the example of looking for a precise move of 100 pips, which is exactly what Rio was talking about. The probability of hitting a target at the next strong price action or Fib. based support or resistance level is much greater than using fixed targets - so there's one good way to create an "edge" for the EA.

 

Emotion in manual trading is generally a bad thing. The more you discipline yourself, the more you can be consistently profitable, however trading then becomes boring and you feel as though you're trading like a "robot" :) Which is where it gets interesting, because if manual traders when at their best are trading robotically, we must be able to code what they do.

 

I think that rather than emotion, it's the subconscious memory store of previous trade scenarios which the skilled trader feels as a "hunch" which is very difficult to code. In addition, our ability to see everything at the same time is very valuable. We can look at a chart and instantly make quite complex approximations to form a view of where price is heading next - this sort of thing can be very hard to code too - that's where the pattern recognition comes in.

[Cyrillic]

Re: Why do most EA's fail ?

 

You people forget that market is chaos because, its not driven by traders consciously but buy large corporations and governments exchanging goods and services around the globe. People that speculate on foreign exchange tend to think thats its a bunch of traders sitting in front of computer all day buy and selling or "big player" or some other nonsense. While its all much simpler, Toyota US has to make a transfer to Japan (do they care what exchange rate is, no, its all business operations, so they convert 1 billion dollars into a trillion yen, BOOM USD goes down yen goes up) Now you got a lot of institutions out there doing that all day this is why price usually moves, not because Warren Buffet says so, not because "big players decide to take money from the market" This market is so big that it would consume anyone that wants to fumble it, unless that anyone has huge capital which they are not afraid to lose and huge i mean trillions, which not a single person we know of has, so big players theory moving market on purpose is almost absurd.

As for gut feeling, we do not know what exactly generates it, for all we know it could be random sound of a dog barking outside which creates that nerve impulse which turns out to be right. Following it a few and it being right it gets anchored in memory as being good, thus profitable decision. Does it always work, i doubt it. People have made Neural model ea's (Russian ea creators loves those) which are overly optimized and slowly learn to fail. All we as traders can do is notice patterns and use them to our advantage, this is why simple ea's tend to work and complicated 10 indicator confirmation abominations fail, market on lower time frames is random, on higher time frames its fundamental, in its highest its a way to see whats happening in the world, in the lowest its a form of gambling and speculation and as soon as you start treating it this way, profits will come.

[soundfx]

Re: Why do most EA's fail ?

 

Hi Cyrillic,

 

Thanks for your comments again!

 

Your first paragraph is interesting because I used to believe that the bulk of FX transactions came from commercials and central bank transactions, however now I'm not so sure.

 

I found this comment on

http://www.newyorkfed.org/education/fx/foreign.html

:

 

"Banks and other financial institutions are the biggest participants. They earn profits by buying and selling currencies from and to each other. Roughly two-thirds of all FX transactions involve banks dealing directly with each other."

 

What's sure is that any one bank couldn't expect to significantly move the market for any length of time. Even central bank intervention has a limited timespan because the market is so huge. However, there are a lot of banks actively trading FX who can get a decent chunk of their overall profits from such trading.

 

So...the market may not move with one big player as for smaller stocks for example, though it's possible that the "group think" of FX traders across the globe causes prices to move in patterns which give rise to some predictability.

 

I agree that "gut feel" is very hard to pin down. However, if we keep saying that these things are too hard, we'll never attempt to even try and approximate them.

 

I like your comment:

 

"People have made Neural model ea's (Russian ea creators loves those) which are overly optimized and slowly learn to fail." =))

 

I've never been a fan of Neural nets and Genetic programming and all that stuff lol. I'm all for keeping things simple and I believe that we can create systems which may appear complex on the surface but they are actually built up of lots of simple modules which when added together creates a whole which is greater than the sum of its parts.

 

You're spot on about price in the monthly and weekly charts being closely tied to global economics and fundamentals and creating an EA to trade longer timeframes like this is pointless.

 

Even though there's so much "noise" in the lower level charts, in theory it's at these speculative levels where the EA should be most effective. EAs could be built to purely work from momentum for example. If a 1 min GBP/USD candle has moved 10 pips (which shows quite strong momentum), I think we can place a trade safely in the same direction and scalp 4 pips on a regular basis. I've not tested this idea, though I think this is another way forward in creating EAs rather than looking for MA crosses etc. and taking losses on false signals all over the place.

 

This thread certainly exercises the old grey matter :)

[scarface]

Re: Why do most EA's fail ?

 

Hi all,

 

Good day,

 

I want to share my opinion as well. Most EAs fail because their strategies don't fit all markets conditions. Take an example of any Martingale EA. This kind of EA is going to make a lot of money when there is a side-way market though increasing the lots every certain position. The problem is that if there is a very strong trend coming up, the result would be blowing up the account.

 

Now take another type of EAs that is based on strong trends, like Hi_Lo EAs. They tend to make a lot of money if they encounter strong and none returning (none reversal) trend. This kind of EA is good, but fails badly in the side-way market.

 

If we combine both ways (types) together based on a smart indicators, we would get a very successful EAs.

 

I hope this is useful.

 

Best wishes,

[soundfx]

Re: Why do most EA's fail ?

 

Hi Scarface,

 

Thanks for your comments :)

 

You've raised a good point about creating a "hybrid" EA which can handle both ranging and trending market conditions.

 

The tricky part implementing something like this working out whether we're in a trending market or not. ADX is the classic indicator which is supposed to distinguish between trending and ranging markets, though I don't think that would be enough.

 

Possibly the key is to look to the higher timeframes to determine the prevailing trend there and even if we're trading a ranging market, we only enter trades in the direction of the higher level trend, in theory that would give us the best of both worlds.

 

Of course, if we're always trading in the direction of prevailing trend at some point the trend will end and we'll have to take a loss. However, provided we have plenty of winners along the trend that would be a small price to pay.

[William1713006271]

Re: Why do most EA's fail ?

 

I've read it at somewhere that the 1st reason most EAs fail are because EAs are created to predict the price (the left over) and not to predict the market (the real movement).

 

Market consists of people (gov, banks, institutional traders, big individual traders, etc) People come and go in this financial markets, even big banks got liquidated. When most top players use some kind of similar strategy, they create the market movement. Each of them has preferred trading strategy. Thus when one of the top player goes out of this business, the market become "chaos" thus the EA strategy will fail. Then another big player comes along with it's holy grail strategy to beat other big players. Thus, the market becomes chaos once again. Thus, EA fails again, coz the algo is changing.

 

e.g. Lehman Bro went to bankrupt, it is considered as one of the biggest financial player. It must have it's own "holy grail" strategy. When it went bankrupt, the market will change.

 

I also read on one of the book, George Soros said that he will only trade when he feels something on his spine. As a big player GS is surely somewhat moving the market. Now, how can EA predicts GS's spine feeling? :D

 

Just my 2 cents.

 

Regards,

[soundfx]

Re: Why do most EA's fail ?

 

Hi William,

 

Thanks for your thoughts.

 

I agree that this longer term change in the overall market can cause big problems with any automatic systems.

 

Typically fixed value stops and targets systems come a cropper when the market compresses or expands from one year to the next. For example, one year 60 pip stops may allow us to ride through all fluctuations in a day's trading in one year and in the next year the market volatility expands in general and those 60 pip stops start getting blown all over the place.

 

It's impossible to attempt to predict what impact bigger players will have on the market, all we can do is design a program which is flexible enough to handle whatever the market is currently doing - hence it will need to have dynamic stops and targets and also to primarily use price action in making decisions.

 

As for GS's spine feeling - that's a really tricky one lol.

[robbye67]

I haven't been active on this forum for a while because I am about to take my option trading company live again for the second time. I noticed this thread and I want to add my recent personal experience to give something back to you guys and gals. In August of last year I started out using virtual money. I started out with $25k and through my experimentatiom with indicators and charting, it went down to 19k. I was then able to figure out the correct indicators to use and I doubled my money to 38K in about less than two weeks. I then went live and I made 5k in about two weeks. Well using the same methods, I started losing real money. I ended the year with a $238.37 profit. I took a long break, analyzed and examined waht happen. Going back a bit, I remember someone selling his tutoring classes say "Indicators don't work". Well he is partially correct. Indicators work only in a trending market. Last year I made money when the market had a definite direction (I made money from Options in both directions - up and down). EA's use indicators to base their decision on what to do. If the market does what it does now, then the indicator will make money for you and then make you go broke. The bottom line is that in this market you have to use a neutral position to make money. When the market starts going consistently up or consistently down, then your EA's will also make you money. Also be careful of your broker. Their commissions and tricks can kill your account. I am not going to name any brokers here; however, the same identical spread from one broker would cost me $268.00 and another broker it would cost me $42 and some change. Finally, if Forex is your game, be aware that it is still the wild west for that opportunity and that market isn't as well regulated as the other markets. Also be aware of your broker. I was using a MT4 platform, when I was considering 4X, and I saw with my own eyes the price of the currency first shoot to the heavens; then shoot to the grave and return to normal. About five minutes later the spikes were gone like nothing ever happen. Good Luck and good trading.