Reversi: a dynamically calculated x-candle reversal EA by Steve Hopwood

Looking forward to giving this one a run Steve, currently set up on 1H charts, using minimum candles 6 (all other settings OOTB).

EA has been trying to place a trade on EURCAD and getting the following error message:

EURCAD OP_SELL order send failed with error(130): invalid stops

No mention of this from anyone else, so I will try downloading a new platform and ea again and see if this resolves the problem or it could possibly be my broker and their spreads

Edit: OK, all fixed, I had to set CriminalsECN to true

Been running v1.04 since Friday night and the results are staggering forward testing. I have it running on a $500 demo account using EU, EG, GU M1 charts (settings below) and account balance is up 68%!!

Just added AU and USDCHF as well to see how they effect the results.

Settings:

DisplayGapSize=30
Lot=0.1
WithTrendLotMultiplier=1.0
ProfitTarget=5
SingleTradeProfitTarget=7
MagicNumber=0
TradeComment=
CriminalIsECN=1
tt=----Trading hours----
Trade_Hours=Set Morning & Evening Hours
Trade_Hoursi=Use 24 hour, local time clock
Trade_Hours_M=Morning Hours 0-12
start_hourm=0
end_hourm=12
Trade_Hours_E=Evening Hours 12-24
start_houre=12
end_houre=24
fsh=Friday stop trading hour
FridayStopHour=24
fta=For the Aussies
rsiin=----Rsi inputs----
UseRsi=0
RsiTf=1440
RsiPeriod=20
rsap=Applied price: 0=Close; 1=Open; 2=High
rsap1=3=Low; 4=Median; 5=Typical; 6=Weighted
RsiAppliedPrice=0
SaturdayStopHour=-1

Oops looking at too many EAs at mo and i must have click reply in the wrong one. I am testing this as well and it shows over the past yr using 99% model quality that is does have legs.

FTA = for the aussies

What is this other EA ?

I'm experiencing the same...

Besides the minimum distance between trades, how about if the trigger for recovery is the same as the original entry? (a same-direction candle run that signals an inminent reverse)

Ok I know it's early but since I've been running the bot on a small TF I was able to observe the bot's behaviour and performance. Here's the setfile I used:


gen=----General inputs----
Lot=0.03000000
TakeProfit=0
StopLoss=0
MagicNumber=0
TradeComment=
CriminalIsECN=0
cst=----Candle statistics inputs----
CandlesInSeries=0
AutoCalculateStats=1
CalculationCandlesCount=2688
UseLongestSequence=0
UseMostFrequentSequence=1
SequenceMinimuCandles=6
ShowStats=1
tt=----Trading hours----
Trade_Hours=Set Morning & Evening Hours
Trade_Hoursi=Use 24 hour, local time clock
Trade_Hours_M=Morning Hours 0-12
start_hourm=0
end_hourm=12
Trade_Hours_E=Evening Hours 12-24
start_houre=12
end_houre=24
rec=----Recovery----
UseRecovery=1
Start_Recovery_at_trades=2
Use1.1.3.3Recovery=0
Use1.1.2.4Recovery=0
Use1.2.6Recovery=1
BreakEvenLineColour=16711680
RecoveryBreakEvenProfitPips=20
UseRecoveryTradeProfitLockin=0
rts=Recovery trailing stop
UseRecoveryTrailingStop=1
RecoveryTrailingStopAt=10
RecoveryStopLossLineColour=255
tmm=----Trade management module----
BE=Break even settings
BreakEven=1
BreakEvenPips=20
BreakEvenProfit=10
HideBreakEvenStop=1
PipsAwayFromVisualBE=5
cts=Candlestick trailing stop
UseCandlestickTrailingStop=0
JSL=Jumping stop loss settings
JumpingStop=1
JumpingStopPips=10
AddBEP=1
JumpAfterBreakevenOnly=1
HideJumpingStop=1
PipsAwayFromVisualJS=10
TSL=Trailing stop loss settings
TrailingStop=0
TrailingStopPips=50
HideTrailingStop=0
PipsAwayFromVisualTS=10
TrailAfterBreakevenOnly=0
StopTrailAtPipsProfit=0
StopTrailPips=0
hsl1=Hidden stop loss settings
HideStopLossEnabled=0
HiddenStopLossPips=20
htp=Hidden take profit settings
HideTakeProfitEnabled=0
HiddenTakeProfitPips=20
mis=----Odds and ends----
ShowManagementAlerts=1
DisplayGapSize=30

I tested on 19 pairs, on M15 (I changed the calculation candles count to be around 1 month on the M15) for 24 hours now, and the results have been around the account up 10% (On a $2000 acct balance with 0.03 lots, NO SL, regular recovery kicking in).

The good thing about this robot is that you get a lot of good entries, it's like "The Beast On Steroids". A couple observations that I leave for discussion with the best interest in getting to a better bot for all of us

1. Sometimes you get a pretty good run of same-direction candles run that could qualify for a "reversal signal", but it doesn't trigger a trade because, on the middle of the run, there is a small candle in the opposite direction that "cuts" the run and frustrates a trade from being taken. I don't know if it's easy, but it might be worth adding a kind of filter (based on the total distance of the running candles) to "ignore" that small opposite-direction candle that frustrates the trade from being taken

2. Also, sometimes you get a pretty good move in one direction but instead of being of say, the 6 same-direction candles required for triggering a trade, it's just 4 of them and then you get a big nice reversal. Maybe the bot could incorporate some sort of logic that lets it trigger a trade despite of not having the "X candle" movement required by the algorythm (provided the movement was big enough).

3. As Steve mentioned before, recovery kicks in too early (although sometimes it didn't trigger at all, so there may be a bug). As I stated in a previous post, I think that besides having a "MinimumPipsForLastEntry" sort of input, recovery trades should be triggered with the same logic as the original entry trades to increase the chance of success.

I know this would be hard to code and a lot of suggestions, but please bear in mind that my intention here is to contribute to getting to a better bot and not just asking and asking without giving something in exchange, so that's what I try to do with this post.

Hi there,

what an interesting approach!
I like statistical analysis too...

I had coded a script primarily for Renko charts that would display statistics of sequences of consecutive up and down renko blocks, to analyse the probability of reversal, but this script can also be applied to regular (non-renko) charts.

Here is my original post:

http://www.forexfactory.com/showthre...12#post4027612

=====

I have recoded it as an indicator, which makes it easier to control its inputs, and will recalc the stats when switching the chart's timeframe.

=====

Inputs :

extern int BarsBack=0;
0 means it will use all the available bars on the chart
any other number is the number of bars back from present to take into account for the stats;


=====

What it does:

It displays 2 columns: UP and DOWN counts of sequences of consecutive candles;
Each row shows the count and probability of a sequence of UP or DOWN consecutive candles of the corresponding number of candles.

Example on EU/H4:
[EDIT: i forgot to put the picture here last night... oops! here it is, i recreated it, and updated the figures below, which are almost identical
The sample contains 5493 H4 bars in total ]

http://www.forexfactory.com/attachme...7&d=1305672250

This tells that there are 1409 sequences of UP candles in the sample,
50.4% of the sequences are 1 bar long : 1 UP is immediately followed by 1 bar Down: it is a coin toss.
what a surprise
25.7% of the sequences are 2 consecutive bars (i.e 2 UP bars followed by a DOWN bar)
etc.

=====

How we can use this information:

we can determine the probability of having a reversal after 5 or more UP bars for instance: it is
100 - 100 * (sum of counts of UP rows with sequence lengths >= 5) / total number of sequences

With my example picture:

prob_of_reversal_after(5) = 100 - 100 * ( 48+23+5+2+6+1 ) / 1409 = 94%

With 4 bars:

prob_of_reversal_after(4) = 100 - 100 * ( 68+48+23+5+2+6+1 ) / 1409 = 89.2%


=====

What it DOESN'T tell you:

although we have over 89% chance (9 times out of 10!) of having a red candle after just 4 green candles, this does NOT tell us HOW BIG this red candle will be...
It may very well be a very small 2 pips reversal candle after 4 big 50 pips candles, who knows ?...

That is why this indicator and information may have a much better use on renko charts: we KNOW that all blocks are of the same size, and therefore we know the exact size of the reversal block, and we can set our TP accordingly.

Another statistical element could be displayed to give us the average reversal candle size and its standard deviation, so we could set our TP according to these values to increase the probability of hitting the TP ?

But it's too late now... I need to get some rest!
Sq

Excellent piece of work Squalou!

Great work, Squalou.

I agree with your assessment entirely. I think this is why we need to use the consecutive candle count to determine the set up only, and then use something else (an indicator for example) to trigger the actual trade. A dynamic TP and SL derived from the typical reversal candle size would also be interesting.

Cheers

Invisible

Fantastic work sir, we are very much in your debt. The problem of not knowing the size of the reversal candle can be solved by Renko.

I was thinking of placing a hedge after a -ve trade then using recovery waiting for the next set up. If there are any mathematicians (or statisticians to be more precise) around maybe they can be kind enough to give us the probability of 5 or 6 candle set up failing 2, 3 or 4 times.

Kind regards

Squalou, thanks a lot for outlining this in such a clear manner, that is exactly what I tried to explain to Steve a couple of posts above...

I have been playing with this for 2 days and....

1. We are looking for a reversal, eg we have 5 green candles and are looking for a red candle. During that up move we should observe a positive value of RoC (rate of Change), where as upon reversal the RoC of C0 should be lower than C1. It looks to me when using a three to five period RoC (given that we are looking at same number of candles) should give some clue whether a reversal is more likely. A more conservative approach would be to only enter when RoC(x)<0.

2. This obviously does not take care of your point that the reversal candle could only be a doji (ie a few pips) and then continue the previous move.

3. I was wondering whether a calculation of average candle size over the last x candles would be meaningful, eg average candle (body) size over last candles is 20 pips, enter reversal only if RoC C0 is less than RoC C1 and reversal candle body is at least 50 or 100% of average of previous x candles size

4. If I remember my maths correctly, a sample of 1024 should give a reliable size for analysis and should reduce calculation load

I added the missing pix to my previous post...

Ok, i've made an attempt to work out the average reversal candle size and standard deviation for each consecutive run of candles in any one direction. I've appended the values to the existing table Number (%) Av pts [StdDev pts].

Not sure it gives us what we are after, but haven't got time to look at it at the moment so if someone else could review numbers and code that would be good.

Thanks for this effort, magft !

I tested it, and fixed the stddev calc.
I also changed a bit the output layout,
Avg and SD values are shown in pips rather than Points.
I added a "Prob" column showing the probability of reversal for each number of consecutive bars, this is the first thing we are interested in.


Here are the columns:

"n": the number of consecutive candles of the same color before a reversal candle;
"nb": number of sequences found in the total sample for this "n";
"%": the equivalent percentage of the total number of sequences
"prob": the probability that a reversal will occur after that number of consecutive candles;
"Body": average size of the body of the reversal candle that follows a sequence of "n" consecutive candles of the same color, in pips;
"SD": the standard deviation of the Body size, in pips;


Now, playing with this tool a bit on multiple pairs and timeframes shows that the StdDev is very large compared to the average reversal body size, and sometimes even exceeds the avg body size.
This means that the reversal body size is widely spread.

Example: EU/H1 shows this:

UP: n=3: prob = 88.4%, body=16.1p, SD=16.1p
(this will depend on how many bars are in your charts, i used 20000 bars in my sample, going back to beginning of 2008 or so)

There is 88.4% probability of having a reversal candle after 3 consecutive green candles, the average red body candle that follows is 16 pips, and 1SD is 16p.

1SD means that there is 68% probability that the reveersal candle body will be in the range [avg-SD, avg+SD], that is [0, 32p].
( you can read more about stddev in wikipedia )

Sooo....
we have 88% chance of having a red candle after 3 green candles, and that red candle has 2/3rd chance of being shorter than 32 pips, with an avg of 16 pips.

After 4 green candles, the probability of a red candle jumps to 95%, with an expected body size of up to 27 pips (avg=14 & SD=13).

On H1.

This seems pretty attractive, doesn't it ?

Now, what is our SL going to be here ?
THAT is THE question.
How much risk are we going to put on the table to reach our 27 pips target ? Or even just 14 pips TP ?

If we set NO SL, then what probability do we have to reach our 14 pips target ?
Much less than 95%, because the SD is large.
We could actually calculate the probability for various TP values based on our sample, giving a 3-dimentionnal array result...

If we put our SL too tight to increase the R:R ratio, there is probably a good chance that the SL will hit before our TP because of the SD being large again -- backtesting would tell us.

If we put our SL far away, then there will be very few loosers (1 loss for 19 wins!) , but possibly with a very large loss -- this all depends again on the statistical average of the 5th green candle if we take the example of n=4.
So we could also evaluate this clearly (yet another column in our results).


If some coders want to add the calculation of the actual win probability for various TP values, please do...
I suggest to add an input to specify the fixed value for "n", and plot the array of resulting win probability for each TP value between avg-SD to avg+SD for instance.

It would be nice to save those results into a CSV file to allow easy handling via excel, much easier than plotting everything on the charts.

And also the avg size of the "loosing candle", i.e. the size of the "non-reversing candle" that we were expecting (the 5th green when n=4).


Another interesting statistical result would be to have the avg+SD of the reversal candle potential, that is: not just the body, but also counting the wick beyond the body, which is what we can expect to reach at a maximum on the reversal candle.
This increases the TP potential.

We could even count the average potential of the total reversal move, not just the single reversal candle.
This would increase the TP potential even more.
But the SD will probably be even wider, degrading even more the quality of the results, and therefore the probability of success of the trade...


Not easy to solve the equation...


The additional stats i suggested to evaluate could sched some interesting lights...


Sq

that result: 1 looser for 19 winners (95% prob) if we wait 4 consecutive green H1 candles...

That is suspicious to me...

Who wants to point me to my mistake ?

Because if that would be true, it would mean that, in order NOT to be profitable, we should have a non-reversing candle of more than 19*avg reversal candle size, i.e. the green continuation candle should be > 266 pips ?!?!?!?!?!?!

Something MUST be wrong here.

Or is this the HG ?...

I will add a Profit counter per "n" value to see the outcome in case the trading rule would simply be
" enter a counter-trade after n consecutive bars ";


Sq

No problem - Steve will be along soon to sort this out for us ....

Here is a version that will calc the profit if the rule would be :

" enter a counter-trade after n consecutive bars of the same color ";

(hopefully the coding is correct for that part at least....)

Now we can play !

First results:

EU/D1 over the last 1000 D1 candles (3 years or so): quite nicely profitable;

And a new rule appears here:

We can CUMULATE the profits by trading TOGETHER ALL "n" values, and BOTH directions !

That is a if we were running many R's, 1 for each "n" value from 1 to 9, and 1 series running in BUY-ONLY, and another running on SELL-ONLY...

Any lower TF seems RUBISH !!


Interesting finding at least !


Macman, thanks to stop by, we neeeeeeeeeeed you....

Sq

Very interesting finding, Squalou. Thank you!

Squalou thanks for the update, i realised i'd forgotten to initialise and reset to zero the arrays correctly whilst stuck in traffic a silly error but i am still learning MQL4 so this is to be expected.

The results are enlightening, for the different tfs there is quite a difference in performance. I'll have a look at testing out different combinations, i think i'll set a forward test of winning combinations either buy or sell and see what happens.

Mike