1 (edited by Freesby 2018-02-16 17:00:58)

Topic: Very bad strategies in real good in FXSB

I've just uploaded some dax and Gold strategies in the repository that show optimal results in FXSTB, both in the 40% optimization period and in Monte Carlo.
There are no ambiguous candles, the data was downloaded with TickStory for dax and Gold (it would be nice to have them in FXTB by default, I do not think they slow down)

The other cross uses default data

Even the statistical values are not bad.
Once put on a demo account on 2 different brokers, they are horrible.

Consecutive losses higher than the statistics.

As I am getting tired of having these strategies, I would like to know;

with what data do they give you?
What are the statistical parameters I have not considered?

How can you avoid absurd strategies where I open an order when the donchian is at 200 shift 197? and you'll see that the ones I've uploaded are all more or less like that.

The result of 3 weeks of work is about 20 strategies selected (ultra selected, not taken at random) and in a month, the one that has made more is -5%.

Apart from throwing all the strategies? Throw FXTB?
We avoid unnecessary chat type need 6 months of demo.
The strategies loaded are crazy and random, no person would put them in real.

Re: Very bad strategies in real good in FXSB

One difficulty may be that the data is from a source other than your broker..

The time on the data bars may be different from thoise of your broker.

Best approach is to use data from your broker real account, even if it takes months to accumulate it.

Re: Very bad strategies in real good in FXSB

You can always prepare the Ea in FsbPro and then backtest in MT4 or Mt5. Once you have the logic developed it is easy enough to reoptimize in Metatrader to be sure the data is correct.

These things are situations that you have to resolve for yourself, ie find a way to succeed.

Re: Very bad strategies in real good in FXSB

Blaiserboy, I did not start yesterday with Mt4, but thanks anyway.
These things, like how optimization in Mt4 were done, I also used several years of historical data from 2 brokers, compared the candles in mt4 and Fxsb.

If you look at strategies, the problem is in the high and super-optimized values of the indicators.

Re: Very bad strategies in real good in FXSB

the problem is in the high and super-optimized values of the indicators

Fortunately this is easy to be checked. Simply click on a numeric param value and roll the mouse wheel up and down, if the strategy balance lien drops after a minimal change of a param, it is a sure sign that the strategy is over-optimized.

I also frequently "normalize" the values of the indicators manually. I mean I tend to put a more round values. for example if a MA period is 59 I try to make it 60.

If you prefer, you can greatly reduce the over-optimization if you turn on the "Use default numeric values" option on.

https://s9.postimg.org/82sn2h9kv/screenshot_158.png

Re: Very bad strategies in real good in FXSB

the problem is in the high and super-optimized values of the indicators

Without seeing the Equity chart I couldn't say for sure -- but I would tend to disagree this is the problem.

If the Equity chart appears linear and there are many trades that result in a slow, gradual increase -- then it is not over-optimized.

An over-optimized strategy would have an Equity chart with one or more large steps that account for most of its Net gain.  These few, large steps from the past are rare events that may never be repeated again in the future.  The intervals between the large steps are losing trades -- and that's more indicative of what the strategy will do in the future.

Also, using the Monte Carlo feature does provide a clue as to how "fragile" the parameters are.

One last note -- backtesting is no guarantee of good trading results.  There can be a big disconnect between backtesting statistics and how a strategy performs in a Demo or Live account.  I think this is mostly due to the data -- backtesting uses bars (which are not real data), whereas Demo and Live accounts use tick data that arrive fast and furious many times per second.

7 (edited by hannahis 2018-02-19 07:47:45)

Re: Very bad strategies in real good in FXSB

sleytus wrote:

the problem is in the high and super-optimized values of the indicators

If the Equity chart appears linear and there are many trades that result in a slow, gradual increase -- then it is not over-optimized.

An over-optimized strategy would have an Equity chart with one or more large steps that account for most of its Net gain.  These few, large steps from the past are rare events that may never be repeated again in the future.  The intervals between the large steps are losing trades -- and that's more indicative of what the strategy will do in the future.


How does an equity curve look like really depends on the nature of your EA.  If one's EA is a scalper that takes in lots of small profit and suffer occasional big loses, then I would expect the equity curve to have many incremental/steps increase and a few big dig in between these small incremental steps.

If my EA is a breakout, the opposite observations will be made, I'll have big/large incremental steps (with each breakout periods) with many small dips in between these big steps.  Hence these large steps may not signify rare events that won't repeat itself (there again, we need to examine how big these steps we are taking about, normal breakout or black swan events).

Likewise a lot of people like to use high win ratio as acceptance criteria to search for EA, there again, we need to understand what are we looking for?  A breakout EA tend to have "lower" win loss ratio (unless one really know how to use filters to eliminate those loses in ranging times).  Cos a "normal" breakout EA trading "behaviour/pattern/equity curve" is one with a big win follow by many small loses, so the equity curve will have big/large incremental steps followed by many smaller loses in between these large steps.  I said "normal" because that's how the market behave, normally a breakout is followed by a number of ranging days.  Hence a "normal" breakout EA without good filters will then behave "normally" with low win ratio.  But not all breakout EA need to behave this manner.  As for me, I will add in good filters and hence managed to reduce those loses in between those ranging and hence my equity curve will look smoother (big steps followed by flat, cos I've managed to reduced loses during ranging times, as shown in the image attached).

https://s13.postimg.org/rl2k28osj/equity_curve.png

So if anyone is searching for a better Breakout EA, one then need to add in extra slots with different LTF (linked them) and then let the Generator search and add in those filters in the available slots.  A "good EA structure" has 2 main components

1) How to make profit (these are your trading rules for entry/exit for your EA to make money/profit
2) How not to lose money (these are filtering rules whereby you try to restrict your EA from entering into false signals that will deplete your profit).

So for those using Generator to search, make sure you have "enough" slots for components (1) and (2).  If you are only relying on components (1) Trading rules to make profit, EA then to have lower PF cos it failed to eliminated false signals and hence gave away those previous profit earned.  Thus my workflow or my EA structure will have these 2 component in place.  And sometimes, I will have these slots and then let FSB optimise them (the results for the latest FSB version is far better than the past versions.  There are more profitable EA from the optimising process compared to what I got in the past).

In conclusion, don't chase after some "ideal" equity curve without understand how should your EA behave in a market situation.  We all agree that we can't win all the time, losing is part and parcel of the game but if we are chasing after a gradual and consistent equity curve, we are lying to ourselves, we are indeed then chasing after and looking for an EA that can keep winning consistently with little loses (probably we are then looking for scalping EA).



sleytus wrote:

the problem is in the high and super-optimized values of the indicators

One last note -- backtesting is no guarantee of good trading results.  There can be a big disconnect between backtesting statistics and how a strategy performs in a Demo or Live account.  I think this is mostly due to the data -- backtesting uses bars (which are not real data), whereas Demo and Live accounts use tick data that arrive fast and furious many times per second.

Exactly, well said Steve.  To believe in backtesting is liken putting a soldier in a simulated training courses and expected him to do as well in a real war event. 

There are issues that aren't "reconciled" in a backtesting methodology using just a particular time chart data bar (unless A H4 time chart gets it's bar data also from the 1min bar data, then one can tell the price movement within a H4 data bar with greater accuracy). 

Price spike in real market can't be translated to bar data.  Can a bar data really tell you the price movement within the bar? Does the price hit the highest price 1st then go to the lowest price or did it go to the lowest price then go move to the highest price? This simple question can make a big difference between winning and losing for a EA in a real market and may therefore provide a possible explanation for the big disconection between backtesting results and live/demo results.

8 (edited by hannahis 2018-02-19 08:58:38)

Re: Very bad strategies in real good in FXSB

Backtesting is a "controlled/limitation" environment.

It is operating under the assumptions of certain limitations in order to produce something "achievable" (not reliable).

Hence, we have limitations such as

1. entry/exit rules by Bar Opening/Closing instead of as and when rules are fulfilled at current market price.  Because FSB is operating under Bar data, not tick data and hence can't calculate the EA's past performance based on as and when (tick data) the market price is fulfilled.  It can only calculated based on Bar Opening or Bar Closing price (OHLC, Medium, Typical Weighted) to compute your trading rules.

2. use of Previous Bar, otherwise how to compare whether price has increase or not (unless one is to compare the price between the current Bar with the previous Bar.  Since it can't track price movement within a bar, it can only track price movements between previous bars and next bars.

3. No point by point (pip by pip) trailing, instead the price is trailed by the next highest Bar.  So it's Bar by Bar trailing.

Thus if after considering all these possible limitations, how could one still expects the back testing results to reflect live/demo results?  How can one expect the backtest results to be accurate (in terms of reflecting accurately how the EA would behave in real market conditions).

By having the above limitations, it becomes achievable to reproduce the "possible" results how an EA may behave in a backtest, past market conditions.  However, do take note, don't mix up "achievable" as "reliable".  Cos in order to create a "reliable" backtest, the backtest must therefore remove those "controlled/limitations" environment so that it truly is a real simulation of how a real market behaves.  Cos as long as we have those "unreal" controlled environment, we aren't going to get any reliable results.

Since FSB can't track price movement within bar, therefore, "price discrepancy" arises.  This is the reason why I only use 1 min time chart because in my opinion, it is an environment/time chart with the least price discrepancy between 1 Bar from the next Bar.  The greater the "price discrepancy" between each bar, the greater the "disconnect" between backtest results with live/demo results. 

That is why Popov didn't implement the pip by pip trailing cos it would be need those extra "data" which will slow down the whole process tremendously and I think most people pc won't have the capacity to handle such brute force.  So between "quick and inaccurate results" and "slow accurate" results...we have to settle with the quick but inaccurate results. 

Somewhere something got to give.  Cos to produce a "slow but accurate results" it would be costly (in terms of hardware requirement and software cost) and then only a few can afford.  So we rather have something (quick but not so accurate results but still can find you profitable EA) than ended up having nothing (cos most can't afford the slow and accurate one, FSB Enterprise).

9 (edited by Freesby 2018-02-20 18:06:12)

Re: Very bad strategies in real good in FXSB

the problem is in the high and super-optimized values of the indicators

Fortunately this is easy to be checked. Simply click on a numeric param value and roll the mouse wheel up and down, if the strategy balance lien drops after a minimal change of a param, it is a sure sign that the strategy is over-optimized.

- Popov,
I did not understand your message,
if you're only referring to the optimization phase, of course,
but if I let go of the Strategy Generator, I can not set limits here,
so I have to make the generator work 10-12 hours, take 10 "good" strategies, maybe with a moving average 100 shift 199, enter more realistic data, and optimize them all.
You will not find any strategy, because they are not a small optimization, from shift 199 to a realistic value.
If the Generator for each indicator, has a limit as in the optimization phase: work less the machine, look for more strategies, and more appropriate to the expectations of those who use it.

If you prefer, you can greatly reduce the over-optimization if you turn on the "Use default numeric values" option on.

It is too strict a limit

10 (edited by Freesby 2018-02-20 18:27:15)

Re: Very bad strategies in real good in FXSB

hannahis,
you are a good person, but we have different ways of seeing things, and above all we have a different pragmatism and synthesis.
I do not find anything practical in your messages, just talk.
Surely I'm wrong, but do not insist.

For the good Steve,
who knows why all the armies do field tests and exercises and the most trained armies with the same resources and even fewer one) always win.

11 (edited by hannahis 2018-02-20 20:10:50)

Re: Very bad strategies in real good in FXSB

Hi Freesby,

No worries, we have a 6yrs gap in terms of our exposure and experience in using FSB.  So inadvertently, we have different perspective and approaches and opportunities to explore the functionality of the software.  To those who don't understand my paradigm (using FSB from a trader's perspective and not so much via statistical approach), my talking sounds like riddles to them and at times even sounds so absurd and nonsensical, that I fully understand. 

If I'm just talking but yet don't have ability to develop profitable EA, then I'm really just merely talking.  But if I am capable to develop profitable EA, then it's really not just mere talking (so rest assured, I'm not talking for the sake of talking, I'm just trying to drop some tips here and there of things that may help others in their trading.  I'm not here to say that I know it all but I'm merely sharing  my own personal experiences cos I been through lots of frustrations and failures over the years (hence, I know what it is like to keep trying and yet not gotten any expected results) and I'm hoping to shorten some people learning curve and save them some time by pointing to the direction of what works for me).

So it's fine if you don't make much sense of what I'm talk (thus you can jolly well discard them as rubbish, nothing personal, no offence).  Likewise, there are many perspectives we have yet to see them all.

We are all here, trying out different ways and methods to search for profitable EA in a manner that best suits our personal preference and thinking patterns.

There are other people (Dave and Footon) who are most statistically incline in their approaches (and with greater experience, veterans) and probably you will find their comments far more useful and practical and I often like to hear their perspectives so as to widen my own and learn from them.

Best of luck to your EA development process.  I hope you will find a workable method that can help you repeat and reproduce successful EA.