Daily Fibonacci Signals

Post Reply
User avatar
snailbeard
Trader
Posts: 615
Joined: Mon Dec 24, 2012 10:54 am
Location: Just above water somewhere between Oxford & Cambridge

Daily Fibonacci Signals

Post by snailbeard »

An EA which was intended to be a stop gap has done better than expected. The original purpose was not to auto-trade but to to produce regular signals for manual analysis.

The basic idea was that to use the previous day as a guide for the current day. The EA calculates a set of levels from the high/low of the previous day. Often the current day follows the previous day. Therefore we have a direction for entries and Fibonacci levels are used as a convenient price level for launching entries. The EA runs in a demo account and produces just over 50% wins to losses unfiltered.

The following chart shows a perfect set-up and sell entry:
gbpusd-m1-Y1M15-Fibo-counter-trade-9Dec2014.png
This crude method ignores the long term direction. Yesterday closed higher so the method only enters buy trades and it does this when the price pulls back and moves up from a Fibonacci level.

Since about half the signals are winnable trades it is useful for pushing these signals to a phone or notepad and then making a manual decision whether to trade or not.

This cuckoo works from inside ZB4 and uses some of its calculations but is independent of ZB4's other entry methods. It was meant to be a stop gap while trying to stop ZB4 from over filtering it's entries but much to my surprise turned out to be too good to be just a signal generator.

After a bit of tweaking and filtering it starting making more wins than losses. The next screen shot is after a very encouraging back test for GBPUSD:
gbpusd-d1-SDIT-example-23Sep2014.png
You do not have the required permissions to view the files attached to this post.
nonlinear
Trader
Posts: 293
Joined: Fri Jun 01, 2012 2:56 pm
Location: Washington DC

Daily Fibonacci Signals

Post by nonlinear »

Interesting, SB. What filter criteria are you using? Are you just buying at the pull back fibo levels when the prior day is up, and selling at the pull back fibo levels when the prior day is down? Which fibo levels? And what is your exit? If you can get a better than 1:1 risk reward it might have more merit.
thomasmore
Trader
Posts: 269
Joined: Tue Nov 15, 2011 10:35 pm
Location: Knokke, Belgium

Daily Fibonacci Signals

Post by thomasmore »

Hi snailbeard,
Are you referring to a existing EA or are you coding a new one for this approach? Looks interesting as I demo-ed it manually.
Ludo
User avatar
snailbeard
Trader
Posts: 615
Joined: Mon Dec 24, 2012 10:54 am
Location: Just above water somewhere between Oxford & Cambridge

Daily Fibonacci Signals

Post by snailbeard »

ludowillems » Sat Feb 21, 2015 9:14 am wrote:Hi snailbeard,
Are you referring to a existing EA or are you coding a new one for this approach? Looks interesting as I demo-ed it manually.
Ludo
ZB4 is my experimental EA which unfortunately has turned into a monster heap of code with quite a bit of unused old code still eating up CPU cycles. Since it already pre-calculates most things that could be useful at a higher level it is easy to add new methods - not so easy to retire redundant code.

The Fib method works well when there are distinct swings but during weak trends and flat periods with whiplash it is pretty useless. It seems in the recent history the market is less trendy than it used to be.

I have been focusing my recent effort on M1 charts since Fibo.s don't help as much when there can be strong moves in any direction.
User avatar
snailbeard
Trader
Posts: 615
Joined: Mon Dec 24, 2012 10:54 am
Location: Just above water somewhere between Oxford & Cambridge

Daily Fibonacci Signals

Post by snailbeard »

nonlinear » Tue Feb 10, 2015 1:04 pm wrote:Interesting, SB. What filter criteria are you using? Are you just buying at the pull back fibo levels when the prior day is up, and selling at the pull back fibo levels when the prior day is down? Which fibo levels? And what is your exit? If you can get a better than 1:1 risk reward it might have more merit.
The basic minimalist method requires the current day's price to be inside yesterday's extremes to begin with.
The second requirement is to think of a preferred direction, again the minimalist choice is to dumbly follow yesterday's direction - works quite a bit of the time even when today turns opposite.
And finally if price crosses a Fibonacci line it is used as signal to enter in the preferred direction.
SLTP are derivatives of D1 ATR and the default is 1:1

Provided pairs trend for weeks in the same direction without frequent reversal it produces a fair ratio of wins to losses but its no use in flattish periods at the daily level and since the aim is to find a few pairs per day to trade this is only profitable with a reasonable trend.

Which is why I have been experimenting with an M1 signal generator which produces occasional entries. The M1 method is agnostic and can't eliminate price action in the 'wrong' direction.

Wrong direction has to be determined at a higher level or two higher levels. I am experimenting with H1 when the market is flattish but deferring to H4. So if H4/D1 is strongly trending then H1 has no influence on filtering signals and is likely to be in opposite phase to the H4 and M1 when the time is right to enter. However, I'm currently interested in flat-trading so this would be when H4/D1 is weak-trending or flat.

The following screen-shots help us discover what could make a reasonable entry signal on M1:

Towards the end of September GBPUSD goes from strongly trending to ranging and so ask can I find mechanical trading rules which work profitably on non-trending days/periods?
To set the scene: here is multi-day chart of H1 bars:
gbpusd-H1-erratic-swings-22Sept2014.png
You do not have the required permissions to view the files attached to this post.
User avatar
snailbeard
Trader
Posts: 615
Joined: Mon Dec 24, 2012 10:54 am
Location: Just above water somewhere between Oxford & Cambridge

Minute by minute

Post by snailbeard »

Zooming in to the 22 Sept 2014 and benefit of hindsight we can see the day had a single dominant direction with no huge reversals. However the slow green down moving average conflicts with the fast up moving averages and so we wait and eventually we see a long running trend. When the fast MAs diverge after converging with slower MAs it is often a signal to re-enter, it works well when there is a longer trend but in ranging markets it is just as likely to reverse shortly afterwards:
GBPUSD-M1-22Sept2014-noisy.png
So we need more clues before daring to jump off the fence.

In this next shot we see that this was a genuine on-going trend which carried on into the following day. The previous chart stops at Fib 100 and in this chart price breaks through and carries on:
GBPUSD-M1-Tradable-swing-22Sept2014.png
So yesterday was plain sailing with no hiccups, so can we expect this new day to go just as smoothly?
As we are talking about ranging days you already know whats coming so here is London's answer:
GBPUSD-M1-Power-Reverese-23Sept2014.png
In this strong reverse the slow green moving average is totally misleading.

It was quite clear that a pullback was developing but it could have bounced at Fib 100. The bounces are still there but small compared to the power of the price action. Once the break of Fib 100 is confirmed it should be possible to enter a sell trade but we don't know if it will stop at the next Fib line and reverse just as fast.

Perhaps there is enough information near the start of the reverse to be in the trade for a better risk to reward ratio. The following chart is a close-up of the minutes before the avalanche phase of the price action:
GBPUSD-M1-exhaustion-reversal-23Sept2014.png
There are clear signs of exhaustion while we see some very large downward 'superbars'. The fast move averages help us to see that price is edging down. Also both the fast and slow ATR indicate rising interest after a significant lull.

Unfortunately, I don't have the patience to watch charts minute by minute waiting for this kind of setup. However, I do have some code to detect superbars although I had not thought to use it in this way and what constitutes a super-bar as opposed to a big-bar and is it worth spending time on this if there is low hanging fruit elsewhere?

Hopefully after more experiments I'll have a better understanding of how to use this minute-by-minute information...
You do not have the required permissions to view the files attached to this post.
User avatar
snailbeard
Trader
Posts: 615
Joined: Mon Dec 24, 2012 10:54 am
Location: Just above water somewhere between Oxford & Cambridge

Circle of Life

Post by snailbeard »

It does seem like I have moved round in a very large circle. My first promising EA was based on a ribbon of moving averages. The idea was very simple: when the MAs spread out after a pinch it is the start of a new swing. During long trends this produced really good results, but when tested over a long period of time there were many occasions when this didn't work and the long-term result was always net losses or negligible profit.

So here I am again with a ribbon of moving averages but from the opposite perspective: how do I prevent entries at the end of a swing and during ranging periods. As you know the market is becoming more and more rangy these days. The Fibonacci method was working great but we seem to have longer and longer periods of ranging and fewer and shorter trending phases or trending is there but weaker. Perhaps it is a consequence of money printing wars or a reduction in global trade but there was no point continuing trading without being able to switch off trading in the weak/non-trending periods.

It turns out that there is a generic way to set up a ribbon of converging MAs and I find H1 is key time-frame if you want to trade daily trades. There is a big catch though which led to a lot of frustration and anger last week.
In this forum we all know that running the same EA on different brokers results in different results. This difference can be amplified if the hour and day are shifted, it is particularly significant if entries occur at prices derived from pivots and Fibonacci levels.

However, I was initially confident that my MAs would smooth out small variations in the data especially as they did not rely on timezone boundaries. So it would not have been good to be within hearing range of my office when I discovered that the same EA was generating twice the profit on FxOpen as it was on FxPro. The path of some MAs were so different that I was on the verge of closing my FxPro account. However, I also tried the same EA on GP and the results were similar to FxPro. So I thought I had developed a generic method but I had actually tuned that method to one particular broker.

So here is the epiphany:

Not only does each currency pair require its own individual parameters but each pair on each broker requires tuning to the data for that particular broker. So if I trade 20 pairs on each of 5 brokers then just one EA needs up to 100 variants.

So each time I make changes to an EA I need to...
:uff:

I have already taken some steps in this direction since I cannot keep track of lots of set files spread across lots of brokers I now internalize the parameters for each pair. It now seems that I would have to add another layer which internalizes the variations for each broker.

However a different approach would be to develop the EA on a single broker, one which is not likely to be trashed by a Central Bank like Alpari was. So perhaps say GlobalPrime is our reference broker, then let the other brokers copy the trades from a single EA. Provided that the differences (spreads and prices) are limited this should work.
User avatar
snailbeard
Trader
Posts: 615
Joined: Mon Dec 24, 2012 10:54 am
Location: Just above water somewhere between Oxford & Cambridge

Recent best results

Post by snailbeard »

Most of my recent development and testing took place using the data from FxOpen and this also occurred with simplest implementation of the MA-Ribbon convergence filter. It produced a 20% net gain on one pair over about 7 months. The entry method can be any entry method that produces regular candidate trades. I now think of my EA as a pipeline in which candidate trades are thrown into the pipeline which flows into the pipe of filters which then either let the candidate trade pass to the next check or throw it out.

So for this test I used an entry method based on increasing (M1 ADX and M1 ATR) which often generates several candidate trades per day. However, this not so useful if testing on brokers like FxPro with limited historical data.

FxOpen results:
gbpusd-H1-MA-ranging-to-ribbon-Nov2014.png
You do not have the required permissions to view the files attached to this post.
User avatar
snailbeard
Trader
Posts: 615
Joined: Mon Dec 24, 2012 10:54 am
Location: Just above water somewhere between Oxford & Cambridge

Disappointing results on FxPro

Post by snailbeard »

On trying to repeat the tests on FxPro I found I couldn't go back as far as September because of limited M1 data:

So in the first run on FxOpen I was delighted with the results:

Total net profit 2175.87
Gross profit 2679.92
Gross loss -504.05
Profit factor 5.32
Relative drawdown 2.63% (297.70)

The lower the draw-down the more overlapping pairs can be trading at once.
The important value is net profit for a particular test period.

After lots of 'tuning' for FxPro the profits on FxOpen are diminished

FxOpen (after tweaking)
=====================
Total net profit 1838.98
Gross profit 2439.28
Gross loss -600.30
Profit factor 4.06
Expected payoff 61.30
Absolute drawdown 15.58
Maximal drawdown 557.86 (4.61%)

The draw-down on FxOpen is now significantly worse after changes.
========================================
For the last few days I have moved the development over to FxPro and have re-tuned the filters. Some losses were due to entries being delayed by several hours. The results for FxPro are better than expected (but over a shorter test period)

Total net profit 876.82
Gross profit 983.59
Gross loss -106.77
Profit factor 9.21
Absolute drawdown 20.90
Relative drawdown 1.37% (146.34)
Total trades 11

So I am wondering what the results will be like if I move development back to FxOpen or over to GlobalPrime...
thomasmore
Trader
Posts: 269
Joined: Tue Nov 15, 2011 10:35 pm
Location: Knokke, Belgium

Daily Fibonacci Signals

Post by thomasmore »

Nice work! I keep reading.
Post Reply

Return to “Automated trading systems”