Topic: Generated Strategies that seem wrong but work well.

I have generated many strategies using the FSB generator and I keep coming up with strategies that give a good equity curve but the strategy just seems wrong. Here is one I just generated.

Strategy name: EURUSD1hourGenerated3
Forex Strategy Builder v2.63.3.0 Beta
Exported on: 20/08/2011 11:02:00 AM

Description:
Automatically generated on 20/08/2011 09:36.
Out of sample testing, percent of OOS bars: 50%
Balance: 35458.43 USD (27/10/2009 20:00  Bar: 9999)

Market: EURUSD 1 Hour
Spread in pips: 20.00
Swap Long in pips: 2.00
Swap Short in pips: -2.00
Commission per lot at opening and closing in pips: 0.00
Slippage in pips: 0

Use account % for margin round to whole lots
Maximum open lots: 20.00
Entry lots: 5.00% of the account for margin
Adding lots: 1.00% of the account for margin
Reducing lots: 1.00% of the account for margin

Intrabar scanning: Not accomplished
Interpolation method: Pessimistic scenario
Ambiguous bars: 0
Tested bars: 19996
Balance: 59718 pips (69718.08 USD)
Minimum account: -333 pips (9667.30 USD)
Maximum drawdown: 8502 pips (8501.60 USD)
Time in position: 27 %

[Strategy Properties]
     A same direction signal - Does nothing
     An opposite direction signal - Does nothing
     Permanent Stop Loss - 750
     Permanent Take Profit - 1450
     Break Even - None

[Opening Point of the Position]
     Day Opening
     Enter the market at the beginning of the day
     Base price  -  Open

[Opening Logic Condition]
     RSI
     [ A ]   The RSI crosses the Level line downward
     Smoothing method  -  Simple
     Base price  -  Open
     Smoothing period  -  2
     Level  -  16

The thing that is strange is that the strategy goes long when the RSI crosses the level line going down.  To my thinking that would not be the correct thing to do but it works, look at the equity curve. I also have another strategy that goes long when the shorter MA crosses the longer MA downward. From years of reading about technical analysis this would appear to be wrong. Does anyone have any idea why these strategies work?

http://img23.imageshack.us/img23/7659/snap20110820111128001.png

Re: Generated Strategies that seem wrong but work well.

Most of the classical theories of the TA was never been backtested. They almost never work as ware described in the books. But the opposite is also true. You can set a classical setup in FSB. Lock it in the Generator and it will make it working by adding additional rules.

Long ago I noticed that the generator sometimes sets "opposite" logic rules for entry or exit. I discussed this with a professional trader. He was not surprised and even he said me that this has more sense then the usual stereotypes.

The true is that  the generator doesn't search a philosophical explanation of the logic. It has both trend following and counter trend rules and searches randomly the best (most profitable) combination.

Re: Generated Strategies that seem wrong but work well.

I have also investigated such an anomaly where the backtest is giving a good results, but the forward testing has been a complete failure. In my experiences, if I generate 10 profitable backtested strategies then only 1 or 2 out of 10 are showing positive results in forward tests. Here is one example:

http://s3.postimage.org/2kdecsyh0/sshot_1.jpg

Re: Generated Strategies that seem wrong but work well.

It is difficult to backtest or forward test any strategy, because the data on demo and real account are different.

EURUSD 5min data from real account:

2011.08.19,19:00,1.43996,1.44055,1.43982,1.44019,410
2011.08.19,19:05,1.44020,1.44110,1.44007,1.44098,358
2011.08.19,19:10,1.44100,1.44132,1.44086,1.44101,275
2011.08.19,19:15,1.44101,1.44107,1.44054,1.44077,239
2011.08.19,19:20,1.44078,1.44078,1.44016,1.44022,259
2011.08.19,19:25,1.44024,1.44062,1.43935,1.43940,404
2011.08.19,19:30,1.43939,1.43978,1.43888,1.43927,347
2011.08.19,19:35,1.43928,1.43949,1.43876,1.43911,322
2011.08.19,19:40,1.43911,1.43922,1.43839,1.43856,309
2011.08.19,19:45,1.43857,1.43954,1.43846,1.43941,323
2011.08.19,19:50,1.43936,1.43942,1.43884,1.43900,224
2011.08.19,19:55,1.43896,1.43938,1.43864,1.43866,317
2011.08.19,20:00,1.43861,1.43909,1.43841,1.43849,467

EURUSD 5min data from the Demo account

2011.08.19,19:00,1.44001,1.44053,1.43983,1.44019,305
2011.08.19,19:05,1.44020,1.44109,1.44009,1.44105,250
2011.08.19,19:10,1.44099,1.44130,1.44085,1.44100,277
2011.08.19,19:15,1.44100,1.44109,1.44056,1.44077,192
2011.08.19,19:20,1.44078,1.44079,1.44018,1.44023,169
2011.08.19,19:25,1.44025,1.44062,1.43936,1.43938,303
2011.08.19,19:30,1.43937,1.43981,1.43891,1.43928,295
2011.08.19,19:35,1.43928,1.43951,1.43873,1.43912,250
2011.08.19,19:40,1.43921,1.43922,1.43840,1.43858,260
2011.08.19,19:45,1.43853,1.43955,1.43847,1.43941,248
2011.08.19,19:50,1.43940,1.43942,1.43885,1.43900,181
2011.08.19,19:55,1.43901,1.43941,1.43866,1.43869,276
2011.08.19,20:00,1.43865,1.43912,1.43841,1.43851,218

The differences are smaller on higher timeframes. There is a EURUSD 1h example, real data is on the left and demo data is on the right.

http://s2.postimage.org/pa2y51d0/sshot_1.jpg

Re: Generated Strategies that seem wrong but work well.

They are not different by much. I would think the differences would be no different from the data from different brokers. If you have built a resilient strategy and factored in some slippage when testing I don't see that slight variances in data would be a problem.

Re: Generated Strategies that seem wrong but work well.

This is whay I always use 1-2 pip slippage while generating strategies. The real problem is that forward testing in demo might be successful, but strategy would fail on real trading.

Re: Generated Strategies that seem wrong but work well.

The difference between demo and real account performance i have noticed, is essentially on stop execution, many more (tights)stops are executed on real trading.

I think it is because of professionals who have access to some order books, and so can see our stops.

This is not as good for large stops.

The better solution would be a "more confidential" stop management,  that would be implemented in FST like an option.The orders (Stop and TP) would not be registered in MT4.

Re: Generated Strategies that seem wrong but work well.

The better solution would be a "more confidential" stop management,  that would be implemented in FST like an option.The orders (Stop and TP) would not be registered in MT4.

The problem in this solution is that it requires working FST and MT4. If one of both stops or internet stops,  there is nothing to stop out our position.

However, FST indicators have access to current position status at the time of calculation and such custom indicators can be build. I'll try to make an example next week.

Re: Generated Strategies that seem wrong but work well.

I completely agree, with this option only there is nothing to stop if there a problem with internet connection or with MT4 or FST.

In this case one solution would be a large "permanent stop-loss" registered normally with MT4, and the "confidential tight stop" registered only in FST, and for each trade separately.

That would be great for high frequency AT strategies big_smile

Re: Generated Strategies that seem wrong but work well.

I return to this topic and no open a new one because it's almost about the same thing "strategies that seems wrong but work.

I 've generated the following strategy with lastest FSB :

Strategy Properties
Same direction signal - Nothing
Opposite direction signal - Reverse
Permanent Stop Loss - 200
Permanent Take Profit - 1000
Break Even - None

Opening Point of the Position
Starc Bands
Enter long at the Lower Band
Smoothing method - Exponential
Base price - Close
MA period - 38
Multiplier - 4,56
Use previous bar value - Yes

Opening Logic Condition
Aroon Histogram
The Aroon Histogram changes its direction downward
Base price - Typical
Period - 180
Level - 0
Use previous bar value - Yes

Closing Point of the Position
Steady Bands
Exit long at the Upper Band
Smoothing method - Smoothed
Base price - Weighted
MA period - 162
Margin in pips - 1738
Use previous bar value - Yes

Timeframe 1min, EURUSD,OOS 30%.

When i have generated it(31/08 at 20:15 EST), the backtest seems to be good with 3000 bars used.

So i have forward tested it(on demo), and it has worked well from 01/09/2001 01:00 EST (first bar in indicator chart).

Today , i have added the newest data, run FSB and accorded the number of bars for the indicator chart to begin at the same time above.

But this time the backtest is bad, even the first trade open at the same time (1:02 EST), the reverse function seems to doesn't work now in the backtest.It's weird!

I just wonder why this happens, if anyone know, it seems like a bug, but with what?

Hope this helps.

Re: Generated Strategies that seem wrong but work well.

It's better to post a screen shot with mark of the "wrong" trades and attach the strategy. The best will be if you post also the indicators values for the last two bars.
When we have this info we can find explanation of the backtest results or to confirm a bug.

We'll be more than thankful if you point out a bug of the program.

Re: Generated Strategies that seem wrong but work well.

Using minute bars, a bar will print even if there is no volume, on some pairs there are many minute bars per day that do not represent activity. Those bars may well distort the calculations.

My 'secret' goal is to push EA Studio until I can net 3000 pips per day....

Re: Generated Strategies that seem wrong but work well.

Blaiserboy wrote:

Using minute bars, a bar will print even if there is no volume, on some pairs there are many minute bars per day that do not represent activity. Those bars may well distort the calculations.

Agree. This problem comes from the broker. Better find data from faster broker.

Re: Generated Strategies that seem wrong but work well.

It must be this, that never happen in higher time frames, the problem appears or not when i modify values of SL,TP,number of bars.

This is why OOS is good when generate a strategy.

Is there any solution to record ticks from MT4 with FST,(working already on tick data?not sure,don't remember exactly) and use it for tick data in FSB?