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

Daily Fibonacci Signals
https://www.stevehopwoodforex.com/phpBB3/viewtopic.php?t=4100		 Page 11 of 21
Author:  snailbeard [ Sun Sep 25, 2016 8:24 am ]
Post subject:  Delayed entries by ATR based filters - part 2
Following on from the previous post, this M30 view of our delayed entry should help us understand why it happened:
gold-m30-delayed-entry-by-atr-filter-10jun16.png
In your view was the entry at A' correct?
Would you have entered?
Unfortunately, we now have a view of the future which can subconsciously bias our analysis.

On one criteria A' is acceptable because price is rising above rising moving averages.
However, there is still significant downward momentum as suggested by the faster blue and red moving averages. So from a manual perspective I would not consider it to be a good entry point.
So A'' is where the filters stop blocking breakout signals. A'' is by itself a good breakout, the problem is that it is a day without much buying interest. Our analysis of ATR is incomplete because another look at it tells us it is not the best of trading days, so we should either not trade or expect reverses to be larger and target prices to be closer. Although the entry for A' was successful in this case there will be many future failing entries, so it might be better to see if we can get a better entry and a more realistic target.

The following M15 chart shows that there were lots of chances trade from a strong breakout:
gold-m15-multi-break-out-chances-10Jun2016.png
From this chart we see that the yellow MA100 is gently rising and provides a rising support line.
What are your thoughts on the strong M15 bar at C'? Would you consider it to be a good entry point?
It seems to tick a lot of boxes and it is only few PIPs higher than A'?
Is there anything about it that would suggest that it isn't a good breakout?
I'll be looking in the log file to see what happened to it.
I have highlighted C'' by a green rectangle not because I expect an entry but because is interesting for different reasons. With reference to to B' and B'' we have rising support over an extended period of time and C" after B'' is a reversal breakout at rising support. Therefore, I am wondering if I should add this pattern as an additional entry method?

Moving on to C''' we can see another missed opportunity to enter just a few PIPs higher than C' so I really need to find out why none of these earlier breakouts made it through excessive filtering.
Author:  snailbeard [ Tue Sep 27, 2016 4:07 pm ]
Post subject:  Golden straddle zone
Sometimes what is obviously a bad time to trade by visual observation can sometimes be non-trivial to quantify in a consistent way. When we are too simple in our assumptions the results often come back to hurt our successful entries else where. We have all tried to quantify ranging by various methods with mixed success. So here is a typical example of obvious ranging for which we need to turn off trading without side effects in other markets:
gold-no-slope-straddle-zone-aug2016.png
On the left we have a well defined slope 'A' where we can acquire pips but prices stall. After a few days the bulls try to restart the upward momentum, but they are repeatedly defeated by bears and eventually we see a new well defined bear-slope at C. Since ATR varies over time and between different assets it can be difficult to find a generic way to consistently quantify ranging. It is easy to detect that the bars cross the moving averages, which happens less frequently when there is a strong trend, but measuring slope strength is more ambiguous.

However, I have tried to do this in relation to ATR, it seems to be helpful for this particular situation:
gold-slope-and-straddle-count-aug2016.png
Looking at the middle of August we can see that the slope values are lower and the straddle counts are higher, so the next step is to try and find a way to use this information without blocking good entries.
Author:  snailbeard [ Sun Oct 09, 2016 4:08 pm ]
Post subject:  Volume based entry filter
Since the analysis to detect price movement and generate signals is now mature, the emphasis is on the pipeline of filters which analyse the other aspects of an assets behaviour. So I recently posted about falling ATR and how useful it can be, but it needs to be overruled when the ATR is cyclical each day and reaches a minimum just before our assets bursts into life again. Just as ATR is cyclical, so is volume, each pair or index or share has its own fingerprint.

If you only trade a few assets you would be happy to enter the times to disable entries manually, but if you want to trade anything which is trending, then it would be a headache to manage hundreds of trading-times.

Firstly, there are the hours when it is possible to trade currency, gold, indexes, shares, commodities, and to some extent these are broker dependent. For example, on FxPro: gold misses a bar at midnight and starts at 01:00, whereas. FxOpen has XAUUSD which doesn't miss a bar at midnight. An EA which is hard-coded and then moved to another broker or to a different version of gold could be using the wrong parameters.

In order to avoid bugs when the EA switches broker or asset type: I try to workout the environment internally first, instead of relying on input parameters. This is quite simple I just look at each hour of each day of the previous week and produce a threshold.

The analysis answers the following question:
Which hour of the day is the first hour in terms of traded volume?
So for the FTSE100 the answer is 08:00 London

but for EURGBP the answer looks like this:

Code: Select all

 HourlySessionVolume: :  
 giH1Vol_FirstOverAvg____ 9
 iH1TradeBarsInOneDay__ 24
 iSymStartHour__________ 0
 giH1Vol_FirstForEntry____ 8
[/size]
So it is recommending that entries are likely to be better after 8 am.

This does seem to have killed off some dubious pre-morning entries.
Using Gold as an antidote to curve fitting for eurgbp and vice versa how do things stand?

I was pleasantly surprised by the improvements to the EURGBP results:
FxOpen-eurgbp-best-NP856-run6Oct.png
So we have a few good trades each month and more wins than losses. There are gaps in each month where the filters calculated it was better not to trade. The challenge is to dynamically adjust these calculations. Having tried to adjust for EURGBP, do the changes spoil our previous successes with gold? Will I be forever going round in circles trying to undo curve-fitting? Before I show you the results for Gold, I can tell you that the current EA version with the same parameters as EURGBP, but used for UK100 hardly trades at all (that might be a good thing).

So here are the results for Gold:
FxPro-Gold-best-NP592-results-6Oct16.png
It would be convenient to say this was all down to the volume filter but it is mostly to do with the careful analysis of price action in the context of ATR.

So for Gold we have too sets of parameters:
the first set relate to volume and the second set to price action:

Code: Select all

01:00 HourlySessionVolume: :  
 giH1Vol_FirstOverAvg 4
 iH1TradeBarsInOneDay 23
 iSymStartHour 1
 giH1Vol_FirstForEntry 3
01:00 HourlySessionVolume: :  
 giH1PAChange_FirstOverAvg 0
 iH1TradeBarsInOneDay 23
 iSymStartHour 1
 giH1PAChange_FirstForEntry -1
[/size]
So the volume parameters tell us that we should ignore anything before 3am CET, whereas the same approach to price action (bug?) is saying price can move any time. I can confirm that there are sometimes big price moves around midnight.

Although I am pleased with the improvements, it is necessary to check if these refinements are common to all pairs or just a few?
Author:  snailbeard [ Tue Oct 11, 2016 12:00 pm ]
Post subject:  No milk today, my trades have gone away
After putting a lot of effort into this EA to be able to trade the UK100 and other indexes, its a little disappointing that the EA hardly ever finds suitable trading conditions. The following chart shows apparently perfect entry conditions, would you agree that this is a day with a good chance of a successful entry?
uk100-absence-of-trade-for-breakouts-20151109.png
The slow moving averages do suggest some sideways movement, but at the same time the mid-speed moving averages help indicate that price is falling, The labels A, B, C and D are where I would expect potential sell entry signals.

The log file is unusually short and it tells me that falling daily ATR is the cause of its reluctance to trade, so reluctant that it isn't even going bother looking for signals. So I am wondering if this is another lagging indicator issue or appropriate caution, looking at the daily bars should help:
uk100-d1-falling-atr-no-signals-11Nov2015.png
From the chart we can see that UK100 is a roller coaster ride. Anyone who can successfully trade through this sticky period is a star, but I'm happy to miss a few good days here and there.

There could be other indexes of interest but for the time being you just can't beat the simplicity of forex.
Author:  snailbeard [ Tue Oct 11, 2016 5:41 pm ]
Post subject:  Relentless
The title could refer to my pursuit of happiness, or a successful EA, but in order to succeed with any EA testing is relentless because you might not be convinced that the EA works well with anything other than the limited test data.

One of the worst aspects of trading are those anomalies created by central banks, fat fingered market movers and high frequency trading algorithms. Many traders were torpedoed by the the Swiss CB and more recently last weeks GBP spike.

I am reluctant to trade an EA live until I'm convinced it won't do something dangerous. Although back-testing has limited value when it comes to unusual situations we might still learn something useful. What would this EA do just before the GBP spike?

I have annotated the following chart where back testing says it would enter a trade and I have no reason to doubt this would have been the result had it been live:
eurgbp-m15-typical-entry-6Oct2016.png
The daily bias is up, an early breakout was filtered out, but after lunch a similar breakout signal is permitted to enter a trade at 14:15. The stealth-take-profit price is beyond this days range, so the trade runs through the night.

Lucky Number Slevin:

For some people the 7th October must have doubled or more their account balance:
eurgbp-h1-overnight-spike-profit-7Oct16.png
The effect of the spike in this case was to sweep past the the stealth take profit price and pass the hard take profit price. However, the back-test is optimistic about how fast it responds to a super spike. In these situations the price often goes well beyond stop loss and take profit. The real profit could have easily have been 10 times higher.

Perhaps some lucky friends in this forum can tell us how far past their take profit price the trade actually stopped? Of course if it was a MM broker you can image it was ended at TP when a profit, but well past SL when it was a loss.

So we have answered the question about spikes but not the one about curve-fitting the data: This is the first time I have used the data of the most recent 4 weeks to test the EA and the results are as follows:

Code: Select all

1	2016.09.12 10:28	buy__	1	0.18	0.84751	0.83259	0.86473	0.00_	5000.00
2	2016.09.12 15:01	close	1	0.18	0.84504	0.83259	0.86473	-55.60	4944.40
3	2016.09.16 13:15	buy__	2	0.15	0.85178	0.83604	0.87012	0.00_	4944.40
4	2016.09.16 20:48	close	2	0.15	0.85683	0.83604	0.87012	92.18	5036.58
5	2016.09.22 10:00	buy__	3	0.15	0.85970	0.84394	0.87816	0.00_	5036.58
6	2016.09.22 20:37	close	3	0.15	0.85670	0.84394	0.87816	-56.09	4980.49
7	2016.10.03 09:25	buy__	4	0.12	0.86913	0.85135	0.89021	0.00_	4980.49
8	2016.10.04 12:18	close	4	0.12	0.87582	0.85135	0.89021	97.53	5078.02
9	2016.10.06 14:15	buy__	5	0.14	0.88152	0.86402	0.90192	0.00_	5078.02
10	2016.10.07 02:08	t/p__	5	0.14	0.90192	0.86402	0.90192	349.46	5427.48
11	2016.10.10 15:00	buy__	6	0.07	0.90083	0.87461	0.93305	0.00_	5427.48
[/size]
This is better than expected due to the anomaly, but even without the anomaly the results are acceptable.
The next step as per the title is more pairs and more testing :uff:
Author:  snailbeard [ Mon Oct 17, 2016 11:59 am ]
Post subject:  optimisation - a fools paradise?
Is there a situation in which it is ever useful to optimise an EA's run-time parameters?

Why am I delving into the murky world of curve-fitting by optimisation?

I scratched my head and thought how do I know what stop-loss and take-profit to use?
Can I think of calculations which will give me an answer?
I already use D1 ATR to give me some kind of ball-park.
Although I could calculate reverses on particular time-frames, they can vary throughout the day, so reverses need to be measured from the moment of entry, not before.

In order to calculate how much room is needed for a trade to not be stopped out there needs to be some knowledge about how far the price can reverse after starting a trade. This will be different for each trade and likely to vary over time. Some assets might not be worth trading because of large reverses.

Initially, I thought this EA was going to be a waste of time for EURGBP, but when I doubled the SL and TP it went from losses to profits. Although, the correct values might not be enduring, I cannot get meaningful profits without finding the correct range.

So grid trading EAs often have ridiculously large stop distances and can make huge profits until a huge spike comes along and wipes the account clean, whereas a cautious EA can only collect a more limited number of pips but hopefully the account won't get burned.

It was also necessary to use optimisation feature for GBPCHF:
gbpchf-optimising-ATR-SL-to-0.5-Oct16.png
For GBPCHF in the table above we see that values of 0.5 and 1.0 give similar results

The ideal profit ratio is when using a tiny stop distance and a large target, but this can only work when there is a strong trend with small wicks. So experience suggests having TP > SL and No. of Wins > No. of Losses.
During enduring trends every man and his dog can do this but for now we must be on our toes.
So I am working my way through the major pairs.
For GBPCHF in the table above we see that values of 0.5 and 1.0 give similar results.
Is there a reason to chose one in preference to the other?
In general smaller values mean trades are shorter. The longer a trade the greater the risk, in terms of exposure to unexpected events, so being a cautious person (usually), the shorter the better.
Author:  snailbeard [ Tue Oct 18, 2016 12:28 pm ]
Post subject:  EURUSD - more work to do
Although I have tried to find issues in the latest EA by testing with different pairs and on different dates, it takes of lot of data to uncover all problem areas. EURUSD caused losses in August, but profits in September were very good. Varying ATR based SL/TP had a significant impact as before but it only moved the results between small profits and bigger losses.

However, on looking at the daily bars the reason for losses in August is obvious:
eurusd-d1-september-ranging-18Oct16.png
It is quite apparent that price is ranging: September was a very sticky month. The convergence of two slow moving averages emphasizes the sideways movement.

How is it that none of the existing filters turn off trading in September?

There are also 17 trades in September which is twice the number I have been getting in a normal good month and I can only attribute that to the frequency of up and down swings.

As we have seen in an earlier post there is already a filter which blocks looking for signals if it can determine poor daily trading conditions.

Would it help to look at the weekly bars:
eurusd-w1-obvious-penant-breakout-aug-sep-oct2016.png
The weekly bars form a triangular price contraction followed by a breakout.
The slow yellow moving average is strongly down, but there is slow red moving average with a more gentle upward gradient, so these two moving averages are converging and form a larger triangular contraction.
There is also a strong white horizontal line acting as a centre of attraction as price swings diminish but continue to swing about this level.

Is it a macro economic level caused by global trade?

The other important features are the lower than average volumes and the continuously falling ATR.
Although, it is easy to combine all these features into a trade blocker, the problem is that the really profitable trades in the first half of October are likely to disappear as well.

Going up a time frame to monthly bars we get an even clearer picture about the oscillation of price about the white line:
eurusd-mn1-12month-ranging-channel-to-October2016.png
Is it possible to classify September as a non trading period without blocking October profits?
Author:  snailbeard [ Tue Oct 18, 2016 3:06 pm ]
Post subject:  EURUSD lines of resistance
Following on from the last post, I show the weekly bars in the context of the previous months range:
eurusd-w1-inside-bars-contraction-aug-sep-oct-2016.png
In this chart we can see the September bars contained within the range of the previous month and swinging about the 50% mark. It will be easy to quantify our September bars as ranging but we might lose some of the October profit in the process.
Author:  snailbeard [ Wed Oct 19, 2016 4:51 pm ]
Post subject:  Fibonacci contraction
Continuing the theme of compressed bar ranges within Fibonacci levels, we can relate the D1 bars to W1 bars and W1 bars to MN1 bars.

In the following chart I have used rectangles to encapsulate August, September and the first two weeks of September:
eurusd-d1-visualizing-D1-to-W1-ranging-sep2016.png
We are trying to reduce an understanding of compressed price movements into Fibonacci context and end up with boolean outcome to enable or disable further processing. We might want to inhibit trading until price crosses the red rectangle. We can see from the blue rectangle that the whole of September is reduced compared to August. The days 12, 13 and 14 of September are completely engulfed by the previous week, so we have a double dose of compressed price movement. Ideally, we want to quantify the whole of September as a dead month in terms of volatility. October starts in the middle range of September and could have been another ranging month, except that price makes a lower-low relative to September and this might be the right time to re-enable trading.

From the log file, we can see that we turned the price behaviour into a few boolean values for further processing:
eurusd-D1BarInsideW1Fib236-table-sep2016.png
Author:  snailbeard [ Sun Oct 23, 2016 4:13 pm ]
Post subject:  volume and price action
In an earlier post I described looking at the average volume and price action hour by hour. This is supposed to find dull periods which are bad for entering trades. There is more than one way to do this, for example, we start each days log file with a hint:

00:00 EURUSD: :
Poor H1Vol Flags[0..23]: 110000011000000000000011 (volume)
Poor H1PA Flags[0..23]: 111000010010000000000100 (change of price)

but that is just a snap shot.
Are these constant or do they vary from day to day or week to week?
From the following extracts we can see that there is some variation over time:

Code: Select all

_272_28092016:H1PA : 111001101000000000000000
_273_29092016:H1PA : 111000011000000000000010
_251_07092016:H1PA : 111000101000000000000010
_287_13102016:H1PA : 101001010010000000000001
_245_01092016:H1PA : 101000110000000000000011
_291_17102016:H1PA : 101000010010000000000101
_264_20092016:H1PA : 101010010000000000000011
_294_20102016:H1PA : 110000001000000010000101
_265_21092016:H1PA : 101010011000000000000010
_279_05102016:H1PA : 010001111000000000000001
_271_27092016:H1PA : 111001001000000000000001
_292_18102016:H1PA : 111000010000000000000101
_250_06092016:H1PA : 111000110000000000000010
_258_14092016:H1PA : 111000110000000000000001
_257_13092016:H1PA : 111000111000000000000000
_274_30092016:H1PA : 111000011000000000000010
_295_21102016:H1PA : 110010001000100000000001
_270_26092016:H1PA : 111001001000000010000000
_256_12092016:H1PA : 011010101000100000000000
_293_19102016:H1PA : 110000011000000010000001
_280_06102016:H1PA : 110001110000000000000001
_263_19092016:H1PA : 100011011000000000000001
_278_04102016:H1PA : 011000111000000000000010
_252_08092016:H1PA : 111000101000100000000000
_285_11102016:H1PA : 111001000010000000000001
_260_16092016:H1PA : 111010010000000000000001
_286_12102016:H1PA : 111001000010000000000001
_253_09092016:H1PA : 011010101000100000000000
_244_31082016:H1PA : 001001110000000000000011
_249_05092016:H1PA : 101000110000000000000011
_267_23092016:H1PA : 111001001000001000000000
_266_22092016:H1PA : 101110101000000000000000
_288_14102016:H1PA : 111000010010000000000100
_246_02092016:H1PA : 101000110000000000000011
_281_07102016:H1PA : 011001110000000000000001
_284_10102016:H1PA : 011001010010000000000001
_277_03102016:H1PA : 110001011000000000000010
_259_15092016:H1PA : 110001010000001000000001
[/size]
These results are unsorted, but there is clearly some variation in the hours which are less active.
All times are UTC Page 11 of 21

Powered by phpBB® Forum Software © phpBB Limited