Contest of Expert Advisors inside an Expert Advisor

Evgeniy Trofimov | 15 September, 2010

Introduction

Dear developers and exchange analysts!

Often, many trading systems (TS) are stable for a specific time period, but further, the balance curve starts moving downwards. Trader disappoints in it and starts inventing new grails, optimizing parameters, etc.

I bring to your attention a tool for performing virtual trades - VirtualTrend.mqh. Using the function I suggest, you can open, close and trail virtual trades.

It has the following useful features:

1. The possibility of creating adaptive Expert Advisors that can enable and disable virtual trading depending on the results of former trades.
2. Rating several trading systems (in this example, 5 trading systems are presented) in the percentage terms depending on their profitability. The competition results in a decision of which trading system is the most preferable for further trading at the real market.
3. The possibility of implementing trading strategies, which use several open trades by an instrument, into the MetaTrader 5 platform where cumulative positions are used (this feature is not described in this article, but suggested as a variant of using).

Let's take a look into the first two statements.

Let's take the Competition_v1.0.mq4 code as an example of adaptive Expert Advisor; there, on the competitive basis, a trade system to be used is determined. The Expert Advisor uses 5 (you can add more if you want) trading systems, which aren't chosen randomly, but on the basis of a more or less stable working on the daily timeframe of EURUSD.

The stability of a trading system is determined "by sight" on the results of tests conducted for the maximum period (all history) in the strategy tester. The obtained diagram of change of balance allows determining the average weighted number of deals that cause a change of tendency. For example, a trading system performs the following sequence of deals: PPPPPPPPLLLLLLPPPPPPPPPLPLLLLLLLLLLLPPPPPPPLLLLLLLLLLLLLPPPPPPLLLLLLLLLLPLPPPLLPPPPPP. The tendency of balance changes after about 6-8 trades. When calculating the rating, it is recommended to set the period of averaging of the balance equal to a half of those trades. That is 3 or 4.

Enabling and disabling the rating system, which helps the Expert Advisor to adapt to the changes of profitability dynamics, is managed via the RatingON parameter. Using it, you can easily find a period of averaging of balance and set it for the Tх.PeriodRating parameter.

Each trading system can be enabled or disabled at your will using the Tх.Enabled parameter, where х is the number of TS.

All the tests of trading systems represented below are conducted at EURUSD (D1 timeframe) during the period from 1999.01.01 to 2010.06.01.

Trading System №1

• Condition of entering: when the fast moving average crosses the slow one (see the T1_SignalOpen() function).
  
• Condition of exiting: reverse crossing (see the T1_SignalClose() function).

Figure 1. The test of the trading system №1 without adaptation (RatingON = false)

The result of testing allows drawing a conclusion that after each 18-20 trades this strategy changes its dynamics of profitability.

That is why the T1.PeriodRating parameter is set to 20.

Figure 2. The test of the trading system №1 with adaptation enabled (RatingON = true)

Trading System №2

• Condition of entering: when the CCI indicator crosses a certain level from top downwards (See the T2_SignalOpen() function).
• Condition of exiting: when the signal appears in the opposite direction (See the T2_SignalClose () function).

During the test for the same period we have found out that the stability is equal to about 10 trades.

Figure 3. The test of the trading system №2 without adaptation

Рисунок 4. The test of the trading system №2 with the adaptation enabled

Trading System №3

The logic of entering and exiting is the same as the one of TS №1, however the periods of averaging of moving averages are different.

A little digression: you can rewrite the entire Competition_v1.0.mq4 and combine it from the strategies that consist of only moving averages with different periods.

I haven't conducted this interesting investigation yet, but if I make it, I'll inform you.

Figure 5. The test of the trading system №3 without adaptation

This TS performs few trades during a long period (almost 11 years). I recommend not using this strategy in real trading since the number of trades in the strategy tester is too small.

I've added this strategy for more obviousness of showing my development of the adaptation. The T3.PeriodRating parameter is set to 2 for this strategy.

Figure 6. The test of the trading system №3 with the adaptation enabled

Trading System №4

I've seen this strategy many times in literature.

It seems attractive even without the adaptation, but using it in trading doesn't require an automated trading, since it's a long-term one.

• It enters the market as three moving averages are arranged in the order and if MACD is higher than the specified level - T4.LimitMACD (see the T4_SignalOpen() function).
• It exits if the price crosses the second moving average (see the T4_SignalClose() function).

Figure 7. The test of the trading system №4 without adaptation

This trading strategy has a constant stability, so the period of processing the data for calculation of the rating should be no less than 20 trades. I set T4.PeriodRating=20.

Figure 8. The test of the trading system №4 with the adaptation enabled

Trading System №5

This TS is developed by me and I want to share it with you in such a sophisticated way.

• We buy if the CCI indicator crosses the specified level T5.LevelCCI from bottom up (see the T5_SignalOpen() function). Set the level of closing trades MyLevel lower than the level of opening T5.LevelCCI by T5.TralingCCI indicator points. Observe the CCI indicator and pull up the level of closing MyLevel with a specific step, in a way to keep the distance between the current value of the CCI indicator and the MyLevel level not greater than the doubled value of T5.TralingCCI.
• Close the opened buy position if the CCI indicator crosses the MyLevel level from top downwards (see the T5_SignalClose() function).

Figure 9. The test of the trading system №5 without adaptation

Such chart can be left without the adaptation, but anyway, set T5.PeriodRating=10.

Figure 10. The test of the trading system №5 with the adaptation enabled

Multisystem Expert Advisor

The working of 5 Expert Advisors with and without the adaptation is demonstrated above.

Now we're going to consider an example of collaborative work of those Expert Advisors:

Symbol		EURUSD (Euro vs US Dollar)
Period		Day (D1) 1999.05.24 00:00 - 2010.07.05 00:00 (1999.01.01 - 2010.07.05)
Model		Control points (a very crude method, the results must not be considered)
Parameters		RatingON=false; FastTest=true; file="virtual.csv"; MinRating=50; _tmp1_="---- Trading system 1 ----"; T1.Enabled=1; T1.Magic=101; T1.lot=0.1; T1.Fast=10; T1.Slow=100; T1.TS=7000; T1.PeriodRating=20; _tmp2_="---- Trading system 2 ----"; T2.Enabled=1; T2.Magic=102; T2.lot=0.1; T2.PeriodCCI=30; T2.LevelCCI=200; T2.SL=500; T2.PeriodRating=10; _tmp3_="---- Trading system 3 ----"; T3.Enabled=1; T3.Magic=103; T3.lot=0.1; T3.Fast=30; T3.Slow=200; T3.TS=5000; T3.PeriodRating=2; _tmp4_="---- Trading system 4 ----"; T4.Enabled=1; T4.Magic=104; T4.lot=0.1; T4.SL=5000; T4.TS=5000; T4.LimitMACD=0.002; T4.PeriodRating=60; _tmp5_="---- Trading system 5 ----"; T5.Enabled=1; T5.Magic=105; T5.lot=0.1; T5.PeriodCCI=90; T5.LevelCCI=100; T5.TralingCCI=100; T5.SL=5000; T5.TS1=5000; T5.PeriodRating=10;
Bars in test	2994	Ticks modelled	219840	Modelling quality	n/a
Initial deposit	500000.00
Total net profit	617173.70	Gross profit	1342671.82	Gross loss	-725498.13
Profit factor	1.85	Expected payoff	2373.74
Absolute drawdown	76798.13	Maximal drawdown	172676.05 (28.98%)	Relative drawdown	28.98% (172676.05)
Total trades	260	Short positions (won %)	126 (29.37%)	Long positions (won %)	134 (33.58%)
		Profit trades (% of total)	82 (31.54%)	Loss trades (% of total)	178 (68.46%)
Largest		profit trade	78151.67	loss trade	-18831.39
Average		profit trade	16374.05	loss trade	-4075.83
Maximum		consecutive wins (profit in money)	6 (89681.19)	consecutive losses (loss in money)	21 (-100325.23)
Maximal		consecutive profit (count of wins)	95057.65 (3)	consecutive loss (count of losses)	-100325.23 (21)
Average		consecutive wins	2	consecutive losses	4

Figure 11. The test of the multisystem Expert Advisor without adaptation

The result of combining the trading systems is the increase of profit, drawdown and number of trades.

Now let's look into the same test but with the RatingON parameter enabled.

Period		Day (D1) 1999.05.24 00:00 - 2010.07.05 00:00 (1999.01.01 - 2010.07.05)
Model		Control points (a very crude method, the results must not be considered)
Parameters		RatingON=true; FastTest=true; file="virtual.csv"; MinRating=1; _tmp1_="---- Trading system 1 ----"; T1.Enabled=1; T1.Magic=101; T1.lot=0.1; T1.Fast=10; T1.Slow=100; T1.TS=7000; T1.PeriodRating=20; _tmp2_="---- Trading system 2 ----"; T2.Enabled=1; T2.Magic=102; T2.lot=0.1; T2.PeriodCCI=30; T2.LevelCCI=200; T2.SL=500; T2.PeriodRating=10; _tmp3_="---- Trading system 3 ----"; T3.Enabled=1; T3.Magic=103; T3.lot=0.1; T3.Fast=30; T3.Slow=200; T3.TS=5000; T3.PeriodRating=2; _tmp4_="---- Trading system 4 ----"; T4.Enabled=1; T4.Magic=104; T4.lot=0.1; T4.SL=5000; T4.TS=5000; T4.LimitMACD=0.002; T4.PeriodRating=60; _tmp5_="---- Trading system 5 ----"; T5.Enabled=1; T5.Magic=105; T5.lot=0.1; T5.PeriodCCI=90; T5.LevelCCI=100; T5.TralingCCI=100; T5.SL=5000; T5.TS1=5000; T5.PeriodRating=10;
Bars in test	2994	Ticks modelled	219840	Modelling quality	n/a
Initial deposit	500000.00
Total net profit	227123.75	Gross profit	388438.79	Gross loss	-161315.05
Profit factor	2.41	Expected payoff	2341.48
Absolute drawdown	10921.17	Maximal drawdown	76482.03 (12.48%)	Relative drawdown	12.48% (76482.03)
Total trades	97	Short positions (won %)	50 (40.00%)	Long positions (won %)	47 (46.81%)
		Profit trades (% of total)	42 (43.30%)	Loss trades (% of total)	55 (56.70%)
Largest		profit trade	71192.28	loss trade	-12680.47
Average		profit trade	9248.54	loss trade	-2933.00
Maximum		consecutive wins (profit in money)	5 (80463.85)	consecutive losses (loss in money)	13 (-50753.48)
Maximal		consecutive profit (count of wins)	80463.85 (5)	consecutive loss (count of losses)	-50753.48 (13)
Average		consecutive wins	2	consecutive losses	3

Рис. 12. The test of multisystem Expert Advisor with the adaptation enabled

The balance line flattened, the number of dramatic drawdowns decreased, the profit decreased 2.71 times, drawdown decreased 2.32 times, the number of trades decreased 2.68 times, the profit factor increased 1.3 times.

In the fig. 12, unlike in the previous tests, the diagram of volumes has appeared. It occurred as the result of the active rating system.

It acts in the following manner - at first, the profits Tх.PeriodRating of all closed virtual trades are summed up and the obtained value is divided by the number of days those trades were performed for. The obtained value is added by the cumulative profit of all open positions and divided by the number of days of their being active.

If the obtained value is negative, then it's set equal to zero. This operation is performed for each TS.

The rating system tells one trading system from another using magic numbers assigned to active trading strategies in the Tх.Magic initial parameters. It searches for the most profitable TS by the maximum value of profit and set the 100% rating for this system.

All the other TS are assigned with a rating relatively to the rating of leading TS. In the end, a table with three columns is created.

Example:

Magic	Profit	Rating, %
1	9	40.9
2	0	0.0
3	3	13.6
4	10	45.5
5	0	0.0
6	17	77.3
7	0	0
8	12	54.5
9	0	0.0
10	22	100.0

Conclusion

The represented methodology doesn't give a 100% guarantee of getting a profit; however, I can tell you that smoothing of the balance curve is assured. Namely - decrease of profit, drawdown and number trades performed are guaranteed! Increase of profit factor - probably.

It's up to you to decide if it's good or not.

The description of the parameters of the Competition_v1.0.mq4 Expert Advisor:

• RatingOn – switch of the rating (if it's disabled, then trades in the file must comply with the real trade)
• FastTest – don't change the file at each change of the array of virtual trades during testing.
• file - file of virtual trading (appears in the TerminalPath()+"\experts\files" folder)
• MinRating – minimum rating (in percentage terms) necessary for opening a real position
• Tх.Enabled – TS switch
• Tх.Magic – magic number
• Tх.lot – maximum volume that used for trading when the rating is 100%
• Tх.Fast – period of fast МА
• Tх.Slow – period of slow МА
• Tх.SL – Stop Loss in points
• Tх.TS1 – one-time trailing stop in points (moves the Stop Loss to a break-even level)
• Tх.TS – trailing stop in points
• Tх.PeriodRating - period of averaging of the rating (number of trades of history)

Links to similar subjects: http://forum.mql4.com/ru/23455