stevehopwoodforex.com
https://www.stevehopwoodforex.com/phpBB3/
Print view

Daily Fibonacci Signals
https://www.stevehopwoodforex.com/phpBB3/viewtopic.php?t=4100		 Page 10 of 21
Author:  snailbeard [ Sun Sep 11, 2016 7:42 am ]
Post subject:  August blues
Was the low volume of trading in Gold in August due to summer holidays or just the market going through a phase? What ever the reason it did not reduce the number of break out signals, but most of these break-outs are too short lived to fit with the automated method. It doesn't mean that one cannot successfully trade manually or use a different trading method. It does mean that my generic ATR based SLTP method which produced very profitable entries in July will all be given back to the market unless I can carefully quantify the August cycle.

The following chart shows the ups and downs of automatically converting signals into trades over a series of months:
gold-D1-3month-M1MomBkOut-run8Sep16.png
You would be right to say that the trading is curve fitted. Although the signals are not, they are generated by breakout momentum. The curve fitting is the process of selecting a set of filters. I try to be as generic as possible about this. For example, most of the filters are concerned with blocking counter trades when there is strong momentum in a particular direction. This has been a consistent benefit for trading based on daily ATR. It has been harder to quantify what is meant by falling ATR and volume in a way which does not result in too few trades. The difficulty is always the lagging nature of analysis, where the market begins a new emerging phase after being dormant or erratic. It takes time for the increasing interest to be reflected in the lagging analysis, so we cannot avoid losing some very goods signals some of the time.

In the chart above the winning streaks represent a small proportion of all days and there are quite a few 'missed' trades to be investigated. These could be due to bugs or limitations of signal detection or limitations in the filters. However, the priority is to see whether or not the same approach can be applied to other pairs and stock indexes, without re-curve-fitting.

While I was analysing the behaviour of momentum break-outs there was an idea about a 5:1 TP to SL short range entry. During a winning trade there are often secondary entry points after pull backs. Experienced traders might be successfully trading these manually. Support and Resistance prices can often be fixed, sometimes they follow a channel and sometimes they follow a moving average. What I noticed was that often a secondary entry has smaller price reversals. Is it possible to pick these out and trade them successfully by pattern analysis? With a 5:1 ratio one can do well on a much smaller win to loss ratio.
Author:  snailbeard [ Tue Sep 13, 2016 8:54 am ]
Post subject:  Perfect Fibonacci breakout
Some members of this forum might remember that this topic started with a simple Fibonacci breakout auto-trader and without filtering it produced lots of entries. At some point I might graft some of that code into the current EA. It is probably time that I revisited those early posts to refresh the interesting ideas that came out of those experiments.

The aim of this post is to demonstrate how the slower time frames can provide a context for for the faster times frames. So the first chart shows M5 bars and accelerating downward price momentum. This analysis does not look at the longer term trends it simply looks for enough price movement and momentum to indicate that there is a short term trend in a dominant direction:
eurgbp-m5-momentum-breakout-1025-03Jun16.png
So A, B and C help establish the current context by the time price re-passes the morning open level we have strong downward momentum but we cannot know just yet that it is a false break out, although keen eyed readers might notice long M5 wicks at support just after D.

So that was the easy part and now comes the harder part of weighing up the wider market context and trying to resolve the conflicting indications. So if you were to looking at the weekly context then this signal is going with the weekly context but if we find that the daily context is up then how are we going to resolve conflicting cycles. The answer seems to be follow the money. perhaps not a very helpful statement as your opponents are doing their best to mislead you, whereas market forces eventually reveal the dominant direction.

Looking at a D1 based Fibonacci set-up we have a pattern which is quite typical and more often than not gives useful information:
eurgbp-d1-perfect-fib-set-up-03Jun16.png
In the above chart we don't need the Fibonacci support and resistance lines to point out the obvious between A, B and C where the pull back is small compared to the upward momentum between A and B.
D is our day of interest, so now we have a context in which to allow buying and suppress selling.

Finally, according to this third chart, if my EA correctly determines the days context we should get a good entry:
eurgbp-M5-fake-then-real-breakout-1535-3Jun16.png
Although we appear to have a perfect set-up, buggy filtering or over-filtering of signals is a constant challenge:
According to the log file the EA was about open a buy trade based on break-out momentum but blocked this break out:
15:31 EURGBP: : detect M1/M5 BUY signal cancelled: IsM15BarOC_BelowFallingSma100=true

This is because M1 has moved so fast in the opposite direction that it is suspicious of the sudden change.
However at 15:45 we have a solid M15 breakout.
Author:  Merka [ Thu Sep 15, 2016 12:32 pm ]
Post subject:  Daily Fibonacci Signals
Hello SB

I guess the curves on your charts are linear regressions.
How are you using these linear regression in identifying the trend?
The idea of Fibonacci indicator on previous day candle is good but you need first a filter to identify in which direction to trade.
How do you decide your trade direction?
Thanks
Author:  snailbeard [ Fri Sep 16, 2016 7:31 am ]
Post subject:  Daily Fibonacci Signals
Merka ยป Thu Sep 15, 2016 12:32 pm wrote:Hello SB

I guess the curves on your charts are linear regressions.
How are you using these linear regression in identifying the trend?
The idea of Fibonacci indicator on previous day candle is good but you need first a filter to identify in which direction to trade.
How do you decide your trade direction?
Thanks
The curves are moving averages at various speeds. I do have some regression functions for the Share's 'End of Day guess tomorrow' algorithm, unfortunately, the spreads and number of shares available means trading them when the market is ranging is not very interesting.

In my early experiments with Fibonacci breakouts, I had a simple method: follow yesterday's direction which can work really well a lot of the time. However, due to market moods swings, this could also result in lots of consecutive losses, but I would like to revisit the method again when I get chance.

Alternatively we can look at several time-frames at once.
Every time-frame has it's own current price direction. So M5 bars might be going down but H4 might be going up. H4 might be going up strongly or it might be going up weakly. Since we are guessing which way price will go next, based on which way it went recently, we will always be wrong some of the time, or more likely we will be right about the direction but wrong about the stop-loss or take-profit. My EA does not generally have a problem picking the correct direction but picking the wrong time to enter turns potential profits into losses. The difficulty here is still following a direction on the slow time-frame when the faster direction is the precursor to a price reversal.

It is not really a question of choosing a trade direction, it is really a question of detecting that a particular direction has faded into noise and turning off trading before all the profits are given back as losses, that is where the analysis is more difficult. If you read my posts about FBATT it is all about only trading assets with strong well defined direction, whereas basket trading might use the same algorithm to trade many assets at once without analysing the strength of price direction.

The next post is about choosing a breakout direction on M15 based on H4.
Author:  snailbeard [ Fri Sep 16, 2016 8:47 am ]
Post subject:  Limiting breakouts by H4
In the following Gold H4 chart we can see several days of good trending followed by deeper price reversals on the 14 June 2016. On ranging days the number of misleading break-outs can increase with more chance of hitting a stop-loss:
gold-h4-up-trend-14Jun2016.png
So how can we find a safe break-out to trade on the 14th June or should we flag it as a ranging day not to be traded?
Author:  snailbeard [ Fri Sep 16, 2016 10:20 am ]
Post subject:  Limiting breakouts by H4 - Part 2
In the following chart: 'A' is the start of the day, 'B' is strong pull-back, then price direction returns to H4 direction but the existing break-out detection finds plenty of break-outs but post analysis filters out a number of potential entries due to the strength of the reversal. The EA finally allows trading at 'C', however, by looking at D1 ATR (1545 pts), we can suspect that we have already used up too much of the daily move and our Tp to Sl ratio would need to allow for a reduced price range:
gold-m15-ranging-day-14Jun2016.png
So although we could tailor Tp:Sl for ranging days this does add additional complexities to the EA, I would prefer the entry filters to be less enthusiastic and let through an entry at say 14:00, which would increase the available range.

So can we address two issues in one go?
a) reduce the number of break-outs per day
b) accept a break-out earlier in the day

The following simple function might help us towards that end. It simply tells us if the bar just closed, closed above or below a sequence of previous M15 bars, where the number of bars can be about 16:

Code: Select all

void update16xM15_breakout()
{
	if( HasM15BarChanged == false )
	  return;

	gbLast16xM15BreakHigher = false;
	gbLast16xM15BreakLower = false;
	gdPrior16xM15PriceHigh = 0;
	gdPrior16xM15PriceLow = 0;

	int val_index = 0;
	double high_val = 0;
	double low_val = 0;

	bool bDebugTrace = false;
	datetime dtDebug = 0; 
	
	//dtDebug = D'2016.06.03 15:30';
	dtDebug = D'2016.06.01 16:30';
	
	if( TimeCurrent() == dtDebug )
		bDebugTrace = true;
		
	if( (DayOfYear() == 166) && (Hour()>=5) && (Hour()<=17)  )
		bDebugTrace = true;

	val_index=iHighest(NULL,PERIOD_M15,MODE_HIGH,16,2);
	if(val_index!=-1) high_val=iHigh(NULL,PERIOD_M15, val_index);
	else PrintFormat("Error in call iHighest. Error code=%d",GetLastError());	

	gdPrior16xM15PriceHigh = high_val;
	
	val_index=iLowest(NULL,PERIOD_M15,MODE_LOW,16,2);
	if(val_index!=-1) low_val=iLow(NULL,PERIOD_M15, val_index);
	else PrintFormat("Error in call iLowest. Error code=%d",GetLastError());	
	
	gdPrior16xM15PriceLow = low_val;
	
	if( gdM15Close > gdPrior16xM15PriceHigh ) 	gbLast16xM15BreakHigher = true;
	if( gdM15Close < gdPrior16xM15PriceLow ) 	gbLast16xM15BreakLower = true;
	
	if( bDebugTrace && IsTesting() )
	{
		DoLogIt( CURRENTPAIR, ": ", "\n update16xM15_breakout: " );
		DoLogIt( CURRENTPAIR, ": ", " gdM15Close: " + DoubleToStr(gdM15Close, Digits) );
		DoLogIt( CURRENTPAIR, ": ", " gdPrior16xM15PriceHigh: " + DoubleToStr( gdPrior16xM15PriceHigh, Digits)+ " gdPrior16xM15PriceLow: " + DoubleToStr(gdPrior16xM15PriceLow, Digits)  );
		DoLogIt( CURRENTPAIR, ": ", " gbLast16xM15BreakHigher: " + IntToStr( gbLast16xM15BreakHigher)+ " gbLast16xM15BreakLower: " + IntToStr( gbLast16xM15BreakLower) );
	}
}
[/size]

This concise function could be the basis of an entry system all by itself or combined with other checks to be part of a composite system. In the next post I'll look at some results.
Author:  snailbeard [ Fri Sep 16, 2016 10:43 am ]
Post subject:  Limiting breakouts by H4 - Part 3
Running the previous function for 14 June produces some interesting results. After extracting the values from the log file I have highlighted the lower lows in red and the higher highs in blue:
gold-m15-16xm15-breakout-log-14jun2016.png
However, as we are only interested in the longer term trend, we can ignore the low break-outs. The first high break-out occurs at 14:00, so what do I need to do to to enable a break-out at 14:00 to get through the entry filters? It is important to remember that higher-highs can occur in flat but noisy markets, so we still need to check momentum.
Author:  Merka [ Fri Sep 16, 2016 5:10 pm ]
Post subject:  Daily Fibonacci Signals
Hi SB and thanks for replying.
I read "minuetto allegro" on one early post.
Are you Italian?
I'll keep reading.
Cheers
Author:  snailbeard [ Fri Sep 16, 2016 6:06 pm ]
Post subject:  Limiting breakouts by H4 - Part 4
Now that we have an additional method of detecting a break-out we can return to a previous chart. I have added light coloured rectangles to highlight the desired entry point which would improve the chances of reaching the target price:
gold-m15-revisiting-an-earlier-breakout-signal-1400-14Jun16.png
Where C', C'' and C''' indicate M15 break-outs and C''' is the late breakout, which results in a loss, while finding the breakout at C' would give a better chance of success.

First we have the results without the new breakout detection (showing a loss):
fxpro-gold-late-entry-results-inloss-1500-14Jun2016.png
After adding the new code we can check the log file to see if the flags are set as expected:
gold-extract-from-log-with-signal-1400-14Jun16.png
I have highlighted the more interesting flags and hopefully the trade can now progress to the target price, which is confirmed by the new results:
gold-16xM15-breakout-better-entry-1400-14Jun16.png
So in summary the direction for trading has never been in doubt, but a strong pull back makes it harder to determine when price is moving with the dominant trend. We have to make a choice between filtering and better timing.
Author:  snailbeard [ Sun Sep 25, 2016 6:27 am ]
Post subject:  Delayed entries by ATR based filters
Although using ATR too cull excessive trading can be useful when the price movements are more erratic, we must be careful not to get carried away. ATR like most indicators is lagging, so as well as blocking some bad entries it can also delay some winning trades.

I have been working hard on two distinctly different sets of prices: Gold and EURGPB, as they are quite different I am trying find a compromise which produces acceptable results for both assets and as a result should hopefully also work for other pairs and stock indexes.

The initial results for EURGPB without any changes resulted in a large number of losses and after 3 months the balance was down by more than 10%. However, one of the most important run-time parameters is the calculation of stop-loss and take profit. No trades will be successful unless these are in the correct range.
These are also the kind of parameters which can change over time so ideally we should up-date these dynamically.

EURGPB tends to be more erratic than Gold and needs more freedom of moment, after experimenting with different values this was the best empirical range possible without any other changes:
eurgbp-test-with-sl-atr-1.0-less-worse-20Sep16.png
Several of the bad entries occurred at times which a manual trader would consider unfavourable either in terms of price direction or a lack of momentum as a result I got a bit carried away with filtering out these poor entries and ended up with unbelievably good results, which now amounts to curve fitting.

In order to undo the curve fitting we need to retest with a different data set. So it is back to XAUUSD and the following tables show how the XAUUSD results have been affected:
gold-results-pre-and-post-eurgbp-tweaks-23Sep2016.png
We can see that several previously good entries have been delayed and turned into reduced profits or losses. However, it isn't all bad news as we managed to pick up a new winning entry.
At this stage we cannot be sure that the filters are wrong, it could be that our entry method is missing some better time to enter or more likely we had a good signal which some other filter has blocked.
In the next post I go back and study the first XAUUSD entry that was delayed.
All times are UTC Page 10 of 21

Powered by phpBB® Forum Software © phpBB Limited