Daily Fibonacci Signals

[snailbeard]

Running the long term trades from May 2014 to May 2015 results in a net gain of 7%.
Most of the gain occurs in short bursts at ideal times, there was a period of 3 weeks when equity increased by 10% and there were two periods in 12 months where equity rapidly increased, but in the first few months there were losses after losses resulting in a draw down of 5%.

There are only 22 trades so I am not sure that the results are statistically meaningful. How would it perform over five years?

The following image shows the expected results if we had started using it just one year ago. However, if someone had given me this EA and promised me it was profitable I would probably have cursed them after the first 5 trades and thrown it in the bin, just before it starts making big gains. I would have been a victim of forward testing tunnel vision:

gbpusd-12months-netPf7percent-Apr-to-Apr2015.png

However, I am still looking at a small sample of trades, if the coming year is more cyclical/erratic, then the results would be worse, but if there are longer trends then the results would be better. Since we have doubts about political and economic stability of the coming year, this might not be the best approach. There is the possible break up of the Euro, growing national debts, and slowing growth and attempt boost exports via currency wars. Therefore I am not that keen on riding several months of draw-down on the hope a good long swing later in the year!

I would like to test this method over five years but I only have a couple of years of M15 data. Since this an end-of-day trading method with large stop-losses some loss of resolution could be acceptable. There are enough H4 bars to test over 5 years. However, I have to change the way Virtual-Pending-Orders work. They are meant to enter conditionally. I have made some changes so that VPOs to turn them into direct orders when back-testing is using higher time-frames.

Before doing a longer back test I would like to see if the draw down in bad times can be reduced or the profits in the good trends can be improved, or if nett profit is increased will that also increase draw-down?...

Summary:
GBPUSD: 12 months, Nett Profit 7%, Draw-down (Abs:598.93) (Max:1667.52 (13.60%))
Limitations:
Must trade complementary uncorrelated assets to avoid margin issues

[snailbeard]

Changing the stop loss behaviour can improve results. The US FOMC spike tripped a number of stops. After improving profit protection: pulling all trades when the first one hits the stop, there is nett benefit:

gbpusd-longterm-Npf1043-dd10pct-run31May2015.png

There are still two losing periods but the middle section holds on to more profit

Summary:
GBPUSD: 12 months, Nett Profit 10%, Draw-down (Abs:679.97) (1272.54 (10.51%))
Limitations:
Still must trade complementary uncorrelated assets to avoid margin issues

As well as looking at the longer term results, it would be a bonus to find ways to reduce draw-down during the bad months.

[snailbeard]

The challenge of locking in running profit is a constant theme. What if you did not know that your EA was leaking profits. MetaQuotes crappy back tester doesn't tell you how much profit you might have made if you had held on to it. So when I glanced at some results indicating a growth of 2% after 3 weeks for one pair I thought that was a good result. However, if you add profit tracking to your EA as I had to do, but this time for multiple overlapping trades not just one trade at a time for other short range trading methods, then we get a better insight into this method.

The following image shows how several weeks of lower lows resulted in 6 overlapping trades, the last one coming at the end of the trend/swing before a significant reverse, so price moves up 3 x ATR before triggering a group exit:

gbpusd-backtest-lost-profit-27Aug-12Sep2014-run1Jun15.png

The following extract from one log file helps me analyze profit leakage, it also tracks several aspects of volume and on-balance-volume on several time frames. It tracks volume rise and fall, as well as above and below average volume:

Code: Select all

2014.09.12 04:00 RC: 5, LSP: 297.70, GST: 291.30, GSP: 456.00, GSL: -164.70, GSCP: 147.73, GMaxPf: 741.86
bD1Sma60Up:0, bD1Sma60Down:1, bPriceReverse:1, bVmaAboveAvg:0, bVmaRising:1, bObvFalling:0, bObvRising:0
ObvDir:  D1: U:0,D:1;  H4: U:1,D:0; 
VmaDir:  D1: U:1,D:0;  H4: U:1,D:0;  H1: U:0,D:1;  M15: U:1,D:0; 
VmaAvg Thresh:  D1: U:1,D:0;  H4: U:0,D:1;  H1: U:1,D:0;  M15: U:1,D:0; 
Max % Pf: 7.35, Curr % Pf: 1.46, Leak % Pf: 5.89

[/size]

So we can see right away that there was a running of growth of 7.35% and that the EA allowed almost 6% of that to vanish, because the method requires large stop-losses. A compromise is required to hold on to more profit without crippling the need to swing.

A first attempt at locking when growth is more than 3% does very well in this example, although it might not use the optimal settings when run over longer periods:

Code: Select all

GBPUSD_gpf.txt:392: Max % Pf: 7.37, Curr % Pf: 6.32, Leak % Pf: 1.06
-------------------------------------------------------------------
GBPUSD_gpf.txt:399: Max % Pf: 0.06, Curr % Pf: 0.06, Leak % Pf: 0.00

[/size]

So setting the allowed leakage to 1% after 3% growth, trapped about 6% growth, but these parameters are currently arbitrary and would need to be tuned.

[snailbeard]

Following on from the 1 month back-test, the 12 month back-test produces a net growth of about 20% with the default settings. Most of the gains are in the first 55 trades, however, the final 20 trades on balance lose some of the gains:

gbpusd-LTJL-May14-May15-backtest-run2Jun15.png

The following daily chart suggests that the retracements are growing longer causing stop-outs more often:

gbpusd-D1-Jul14-to-May15-tail-swing.png

Is there a way to recognize that conditions have changed in order to adjust parameters or switch off trading?

[snailbeard]

After a brief exploration of long term trades its back to daily signals. GBPUSD has been swinging nicely for a couple of months - not so good for the long trade but the SDIT (Short Duration Independent Trade) method works well on these clean swings:

gbpusd-D1-AprilMay2015-good-trading-conditions-run3Jun2015.png

In comparison to the long trades, the draw-down here is less than 2% giving plenty of margin for running lots of simultaneous trades on other assets. SDIT was intended to get some trades out of weak trends, so when momentum picked up several of the range filters were kicking in. I have added compensation to adjust the range filtering when momentum is strong.

There will always be a losing trade at the change of direction. However, we can look for clues at the end of a swing...

[snailbeard]

The jury is still out on how reliable OBV divergence is in the long term.
There was a change of direction between 1st and 2nd of June for GBPUSD, which naturally led to an end of swing loss.

In this particular case: divergence on H1 turned out to be a leading indicator. The following screen-shot shows the previous two days OBV had higher highs, while price had lower-lows. Meanwhile the EA had a green light for selling which hit the stop-loss:

GBPUSD-H1-ObvPrice-Divergence-01Jun2015.png

[snailbeard]

Those previous results were on GlobalPrime, running the same EA on FxOpenEcn still produces profits but we are back to the delayed entry issue:

For example on Global Prime we get this good entry at noon (BST):
Sell 2015.05.28 12:01:00 13 0.16 1.53322

whereas as on FxOpen the entry is delayed too long:

31 2015.05.28 14:10 sell 11 0.25 1.52748 0.00000 0.00000 0.00 10836.62

The two hour delay is enough to turn a winner into a loser.

According to the log file I might have shot myself in the foot by using OBV divergence or perhaps it's the time frame:
12:00 GBPUSD: M1 SDIT: Drop by bObvUpM180DivergFlag

Looking at the corresponding M15 chart does not indicate divergence of price and OBV.
I don't recognize the code or understand it, perhaps it was written by Tyler Durden.
Therefore, I have just written a simplified divergence checker...

Re-running on both brokers is now giving surprisingly similar results

Code: Select all

GlobalPrime:
Sell    2015.05.28 12:01:00  13      0.16  1.53322
FxOpenEcn:
0   Sell    2015.05.28 12:01:00  13      0.11  1.53323   

What are the chances that we get similar results on FxPro?
This result from FxPro is too good to be true:

Code: Select all

37	2015.05.28 12:01	sell	13	0.16	1.53306

This is a rare event like a blue moon. However, it is only one pair and only for a couple of months, normal inconsistency will be resumed in due course.

[snailbeard]

After getting very good results during good trends, I went back to the period from September to the end of 2014. This period has a clear trend down in the long term, but there are regular multi-day pull-backs which cause frequent losses. No single filter can turn off trading without doing more harm than good. It was necessary to create a new set of rules. i.e. composite rules combining elements from several time-frames as well as OBV, ATR. The aim is reduce the number of hard filters and make them more conditional.
If this works I shouldn't need to tune the EA to each currency pair, instead the EA should decide whether or not a filter should be applied.

The following chart shows a steady down trend with regular pull-backs:
One can code an EA specifically for this period but then it would do badly when the market changes in the following period.

gbpusd-D1-coping-with-deep-pullbacks-Sept-to-Dec2014.png

This period was initially producing net loss of 3% due to a long series of losing trades.
However, the report shows a reasonable profit without using super-sized stop-losses so the draw-down remains within my 3% target:

Code: Select all

Initial deposit	10000.00
Spread	10
Total net profit	587.01
Gross profit	1149.82
Gross loss	-562.81
Profit factor	2.04
Expected payoff	58.70
Absolute drawdown	74.92
Maximal drawdown	317.32 (2.97%)
Relative drawdown	2.97% (317.32)
Total trades	10
Short positions (won %)	10 (50.00%)
Long positions (won %)	0 (0.00%)
Profit trades (% of total)	5 (50.00%)
Loss trades (% of total)	5 (50.00%)
	Largest
profit trade	300.06
loss trade	-126.22
	Average
profit trade	229.96
loss trade	-112.56
	Maximum
consecutive wins (profit in money)	2 (450.69)
consecutive losses (loss in money)	2 (-240.21)
	Maximal
consecutive profit (count of wins)	450.69 (2)
consecutive loss (count of losses)	-240.21 (2)
	Average
consecutive wins	1
consecutive losses	2

[/size]

[snailbeard]

Could this be the final version?
It is producing the best results so far over the previous few months as well as going back 12 months.
The last few iterations have been about not trading during parts of the year when conditions are less favourable to success. This approach means it is hanging on to the profits gained when trends are well defined.

Although, ATR, ADX and OBV on D1 are pretty slow, combining different constraint rules has disabled periods of the year when direction is almost a random walk. I have given up with short counter trades and concentrated effort on trades which can complete within several hours to several days.

The following graph shows that long periods of losses have been avoided, with some rapid growth during strong trends. The results are better than the long term stacking of trades and avoid the hazards of deep draw-down:

gbpusd-12months-Np17-Dd3-run2015-06-11.png

The summary is as follows:

Code: Select all

Initial deposit	10000.00
Spread	10
Total net profit	1747.82
Gross profit	2109.08
Gross loss	-361.26
Profit factor	5.84
Expected payoff	109.24
Absolute drawdown	57.21
Maximal drawdown	315.24 (2.74%)
Relative drawdown	2.74% (315.24)
Total trades	16
Short positions (won %)	6 (100.00%)
Long positions (won %)	10 (70.00%)
Profit trades (% of total)	13 (81.25%)
Loss trades (% of total)	3 (18.75%)
	Largest
profit trade	282.78
loss trade	-128.74
	Average
profit trade	162.24
loss trade	-120.42
	Maximum
consecutive wins (profit in money)	9 (1443.62)
consecutive losses (loss in money)	2 (-244.35)
	Maximal
consecutive profit (count of wins)	1443.62 (9)
consecutive loss (count of losses)	-244.35 (2)
	Average
consecutive wins	4
consecutive losses	2

[/size]

Note that the draw-down has been kept under 3%.

It would be great to believe it is finished but there is always something that needs adjusting...

[xfactornos]

There is no such thing as a final version Good job at keeping at it! Most people are lazy or run out of patience and give up.