Probably my opinion is very opinionated but I think that the most common and simpler indicators work the best.

Let the software generate the strategies for you -- and then select the ones with the best statistics.  In a few hours it can generate millions of combinations and perform the back testing -- all you have to do is then select the best ones.  Let the software do all the hard work and don't worry too much about ingredients.

Yes -- welcome to the club...

The software can generate, optimize and back test strategies as good (or better) than you can by hand.  But there is no guarantee of their performance in a live account.  I think most everyone has experienced that.  There are other things going on in a live account that we do not have control over -- e.g. the incoming data.  Some people recommend using your broker's data when generating, optimizing, and back testing strategies -- suggesting using your broker's data will yield results more similar to what you can expect in a live account.  Others have suggested using longer time frames -- e.g. >= 1 hour -- since the data will be less prone to noise and false signals.

The problem you describe also occurs when hand-coding strategies.  Prior to FSB and EA Studio traders lost money in live accounts.  Losing money in live accounts is caused by things we either don't fully understand or do not have control over.  Even people who fully understand the intimate details of how each indicator works are not guaranteed of winning.

FSB Pro is used to create many fine strategies -- better than what you could create by hand-coding.  But consistently winning in a live account is something I think we all wrestle with.

Hi Araza,

There are basically 2 groups of FSB users.

1. Let the software does it's work
2. Input your own trading theory and use the FSB to fine tune or search for a better setting/combinations (Smooth, Signal, Base setting).

There is no right or wrong path.  We tend to make our choices on the following

1. The easiest path (does that mean, the best outcome?)
2. The trading knowledge path (figuring out a trading plan)
3. The most predictable/duplicate outcome path

As to date, I haven't heard of anyone who has finally found a successful workflow using the easiest path (maybe someone did and hence doesn't need to be active in the forum cos too busy with his own success).  A successful workflow whereby if I were to follow all the steps and procedures, the final generated EA list will contain a certain percentage of winning strategies in live/demo/nano account.

I wonder anyone has a proven workflow that can yield a certain percentage of success rate.  So that every time I ran another batch of generated EA, with certain data horizon, with certain OOS % and certain MM or MC or whatever the workflow is, I'm sure that I'll get for an example 30% success rate, etc.

If anyone has found such successful workflow, please share with us your successful workflow or concepts behind your development process.

As for me, by default I choose to "trader's knowledge" path many years ago.  Reason being I already have a trading plan that I want to execute.  Secondly during the early years of FSB, I don't find any success in generating etc (probably due to bugs and how the backtest handle previous bar value, etc).  Hence, I choose the trader's input path cos I feel I can get better outcome than the generated path.  FSB has gone through numerous update and thus I've yet to find out whether with the new update, I'll get better results from the generated one.  As I'm too busy with my current work path.  I don't have the time or resources to venture the "easiest" path.  I'll be happy to hear from others who have succeed and would like to share how to replicate this success again and again.

That's why out of curiosity to see how well FSB can churn out EA, I decided to do a mini experiment (before I ever venture to this easiest path) and download EA made by FSB Online Generator (this product has been removed from website).  Currently this batch of EA is still undergoing demo testing.  The initial results was 4 good and profitable EA out of 49 (nearly 10% success rate).  I've created a 2nd batch with added trialing.  Now I've about 20 good and profitable EA out of 98, that's about 20% success rate.  I must say that if I'm a new user, this is a very tempting path to take (why not, with such minimal effort, I get 20% success rate).  At such success rate, I would probably choose this path.  I don't know why this product has been removed.  And I'm not sure how is it different from EA Studio's generator or are they the same.  If they are, are users experiencing such consistent success rate?

Another reason why I stick to this trader's path is because I feel I'm in control of the outcome.  I heard people sharing their tried the "easiest" path but without any much success and somewhere along the line, they feel very "lost", out of control as if what else can they do to make it right.  I don't like those feeling, it's frustrating for me.  Whereas for the Trader's path, I feel I can create a certain percentage of success.  At the end of each workflow, I'm "guaranteed" a certain number of profitable EA.  It's nice to have an expected outcome.  And I feel I'm in control because i can observe the trading patterns and see what new rules I can add to improve it further.  Yes, there are lots of trial and errors too but at least I know my efforts are never in vain.  That is why I like to take the road less traveled.

PS:
If I were to generate EA, I would try to demo test as much EA as I can from the list.  Reason being, the top EA may not be the "best" selection.

It's like a bell curve distribution.  The tail end of each of the distribution may represent either "off sync" EA vs "curve fitted" EA.  Hence my hunch is that those in the middle are the ones that hit the "optimal" setting.  Neither off sync nor over curve fitted.

Algorithmic trading is relatively new. I do believe there are probably people out there who have developed a structure and workflow that will allow them to generate randomly and achieve good results. However, I also believe these same people have:

1. Put a LOT of time and effort into finding this workflow, and
2. Put a LOT of ongoing effort into the testing and maintenance of their generated EAs.

I personally don't believe there will be an ultra-simple workflow that will allow us to press the START button and with minimal effort obtain a whole bunch of awesome, profitable strategies.

To develop a sufficient workflow, we will need to work as a community in a methodical way. "Trial and Error" just won't cut it, because the data we are working with is just too random and our sample sizes are too small (relatively speaking).

It's very easy to come up with all sorts of AWESOME looking strategies with EA Studio/FSB. But the problem is that the software is looking through so many strategies, that it WILL continually find patterns where there are no patterns. Or perhaps portions of these EAs are valid, but some of the indicators are simply noise.

How are we to tell the difference between a "valid" and profitable strategy (i.e. one that generates profits in the future predictably) and a strategy that was developed on a false pattern that doesn't really exist.

As a (admittedly stupid) example, consider the fabulous "when it rains in Kansas City" indicator that I've just developed. This indicator reads "TRUE" on days that it rains in Kansas City and "FALSE" on days that it does not. You are super-excited about this new indicator and fire it up in FSB, using it to generate in conjunction with all the other available indicators.

And...OMG...it generates 100 VERY nice looking profitable EAs within 24 hours.

But something surprising happens when you put these EAs on your live trading server for four months: 80% fail to make money! And they looked so good...

But wait, you say, I still have 15% of my strategies left over! That's awesome! If I can generate 15 winning strategies in a day like this, there is not limit to the wealth I can gain in one year!

You might say, well that's just stupid. That's a totally made up example and it's blatantly obvious to any idiot that this worthless indicator that you're touting isn't going to work worth a damn. And you would be 100% correct.

And, you might say, there's NO WAY that 20% of those EAs you generated would be profitable over the span of four months. I mean, if they are making a sufficient number of trades, that is...say, at least 40 trades.

Aha, there you are WRONG. It is entirely possible, and most likely probable, that 20% of those strategies would show profit over four months. But wait, this is basically a totally random indicator. The FOREX markets don't care if it rains in Kansas City; they won't be affected by that event one bit. No, they won't, that is correct. However, the indicators that they are combined with are not totally random and will yield "some" counterbalance to this stupid indicator.

So why not just remove this indicator that we were so excited about and once again just rely on the default indicators? Well, yes, that is certainly better. But how much better? I would venture to say you would increase your success rate to about 25%. So is 25% "better"? I would say, not really.

Yes, I don't have hard data...but 25% success, 35% success, there really is little difference. The problem is: how are you to know WHICH 25% will perform profitably during that four month span? Unless you develop a better methodology, you won't. It's impossible.

So why not wait and see which ones are profitable after four months, and VIOLA! You have your winning strategies! This might be part of the equation, but it's still not nearly sufficient, as some (many) of those remaining strategies will still be randomly profitable and not viable for future profit.

So just weed them out as they prove that they are not profitable. Again, I am quite certain (no, not 100%...) that this will not work without more in depth workflow. At any given time, you will not have enough (future) profitable strategies remaining to offset the losses you will incur with the ones that will lose money.

As Hannah said above, if the truly "simple" method of EA generation and deployment worked, we'd have more than one rich person bragging about their accomplishments around here (or on forums of similar products). Yes, it's true that some would be closed-mouth, but they would likely be the ones who are much more methodical and thoughtful about the way they approach the softwares--someone who has developed a workflow with much concerted effort.

There are so attempting the "easy" methods that SOMEBODY would have found the "magic bullet" by now.

And, no, I'm not attempting to discourage anybody from using the software. I'm just saying we need to be digging a lot deeper than we are with a methodical approach.

Also as Hannah says, I doubt that anyone can demonstrate a repeatable "success" with simple generation methods. And I think we can all agree that "success" means dependable profits. (Actually, that is not entirely true...autotradingacademy.com is getting rich off the notion!)

I know of one trading group who has done quite well in the last year+ with another software (inferior to FSB). But this trading group has added a LOT of proprietary code, programmed extensive analysis tools, and developed a workflow that they are (reasonably) unwilling to share with the rest of the community.

I thank you all for your detailed and eloquently worded opinions. You have no idea how grateful I am to hear such points of view as I am learning so much from my seniors.

I will continue to bother all of you to learn further and share my experiences along this journey

Hannah
Looking at moving averages....hopefully will get something out of it.

Regards,

Hannah
Like I said earlier that I am really learning a lot form your posts as they dont leave anything to imagination..very detailed and explain everything nicely..I follow you other posts as well and appreciate the time you take to draft them.
I would hope that you would continue to give your opinions..I am sure that it is valued and I am sure you also appreciate people challenging your approach to trading.
Forums such as these are meant for such discussions.
Thanks

Data is changing all the time but the whole idea of data crunching is to discover whether there are any Repeatable Patterns.

X = y + 5.  If X is my age and y is John's age, then no matter how much John's age (data) change, one can know my age, X. 

Likewise, our algo EA is to determine and find such repeatable patterns so that no matter how price changes, there is a relationship or probability of where the trend direction will go.

If data is constantly changing and yet the software doesn't have capability to spot repeatable patterns from these raw data, then what's the whole point of putting our data through vigorous testing only to find out what worked in the past but couldn't identify any repeatable patterns?

Secondly, for the past 90 years people are still clueless?  Can't be, haven't we read many successful stories of companies who employed Phd and mathematicians, what for?  These people can see/spot "relationships" behind numbers.  Does that mean they will share their secret formula so that the rest of us will stop fumbling around and remain clueless.  Well, they probably would be happy to let you remind clueless cos your loss is their gain.  Forex is a zero sum game. As much as I would like to shorten a person's learning curve but will I then share with another person my successful EA so as to stop him from being clueless?  To those who are clueless, they will remain clueless.   But to those who believe there is a relationship pattern behind these data, they will find what they are looking for.

Thirdly, data are just data, they don't carry any rules with them, they can't think and make decisions.  It takes an intelligent master mind to come up with a concept of organising/crunching these data to search for repeatable patterns.  So how successful a software is depending on how the developer design ways to crunch these data into meaning relationships.  People just buy software and expect it to work.  For me, now before I buy another software, I write to the people and ask them more about how they handle their data and do their backtesting.  I question the working and methodology behind the backtest to know whether it will work.  No one seem to ever ask that kind of questions.  Not all backtest tools are designed the same.  It depends a lot on the developer's trading concepts too.  That's why software that are designed by traders will have that added advantages compare to one purely developed by a computer scientist.  A software is only the Brain child of the developer.  So it's a human telling the computer what to do with these data, hence is such person perfect, is he a successful trader to know what kind of rules to apply for better quantitative and qualitative search?  Likewise, how the software optimise the EA, I'm curious too cos I've my own concept of what makes a good optimisation tool.  What make us think that the developer's concept is better than ours?  What's the rationale behind the optimising concepts? Is such method really the best way to optimise an EA?  That's why I question so that together we can test out each other's theory and concepts and hopefully refine the procedure better for greater success.

Hence the whole idea of this software is to let the software apply some rules and see whether given such rules, the EA will be profitable in the past and if such rules are dependable and repeatable patterns of profitable and not due to luck or random chance, then we have greater success rate.  That's why I choose the trader's path because I can input my own version of "repeatable" patterns (instead of letting the software search for me) and hence have good success rate.

I'm not saying the FSB can't generate or find for you repeatable patterns (I believe it can, otherwise why would I be doing the FSB Online Generator project).  I'm saying that since I already found my own repeatable patterns, I don't need to rely (re-invented the wheels) on FSB to search for me anymore.  Similarly, if those who wanted to try to input repeatable patterns (instead of keep doing random searching), they can then try using the simple MA concept as their 1st repeatable patterns.  I raise this up because some people think simple MA theory is too simple it may not work/profitable.  But I'm sharing it worked for me (but does that mean I have to upload my EA just to prove my point? I'm not going to give my "treasures away but I won't mind pointing to people the direction where the treasure is hidden), hoping that people don't find it daunting to try out this trader's path for lack of any good trading theory.

Lastly to those who are testing out your FSB EA during this August, just to give my own (limited) observation, if your EA fail please don't think it means it is a bad EA.  Don't discard your EA prematurely.  It's a technicality issue vs fundamental issue.

Our EA are based on technical analysis.  Hence it doesn't take into account (to certain extend) fundamental issues that went against market forces.  However fundamental issues does affect the price volatility and prices changes.  And because of the prolong fundamental issues to suppress the USD and not like the usual market demand and supply determine the price levels/cycle, thus such external intervention (other than market forces of demand and supply) is "disrupting" the "normal" behaviour of our indicators. Causing many if not most (laggy indicators) to constantly enter wrongly.

Instead of the usual peak and trough, we get unusual and prolong peak (for EUR against USD) and normally when H4/D1 indicator changed, it would have been a "reasonable" trend that bring in good profit but because of the intervention to suppress the USD, such trend change is only short lived and for laggy indicators, it resulted to "false" signals. 

From my observation, I decided to change my approach and now developing shorter term EA so that it will stay "relevant" at all times.