Home › Forums › Development › Two possibly interesting ideas
Tagged: cycle time, Moving Average, tick speed
This topic contains 23 replies, has 6 voices, and was last updated by simplex 11 months ago.

AuthorPosts

Hi folks,
during my holiday, away from trading platform I came up with two probably interesting ideas which I want to present and discuss with you.
 As many of you did, I also read all of Eurusdd’s and CrucialPoint’s posts at FF (or Babypips, where Eurusdd’s name was Certainty). Both authors explain in detail the importance of moving averages in their trading. While CrucialPoint seems to use MAs to determine recent trend, Eurusdd applied MAs to identify the overall cycle length of FX markets like EURUSD. My idea is mostly based on Eurusdd’s concept. He mentioned that we can use MAs for instance to determine optimal settings for the stochastic indicator. Optimal settings may have reached when the slope of a MA becomes flat. Thus, the most representative cycle length for a particular time point (candle) can be identified by minimizing the sum of squared slope differences. As soon as we end up with an optimal cycle length vector (for a time range = multiple candles), we can subtract MA values representing this vector entries and repeat the whole procedure as many times as we want. Thus, finally we obtain a set of optimal settings which when combined may deliver a more suitable dynamic setting for common indicators (like stochastic, RSI, etc.).
 The huge impact of market makers (big players or whatever you want to call these guys/gals) on prizes is usually accepted as main reason for prize movement among traders. Moreover, many would agree that whenever a big order gets filled, the prize reacts immediately which leads to a fast drop or rise. If we further assume that market makers can control markets to a certain degree and manipulate prize to their favour, we can profit from that behaviour by following them. The easiest way to do so could be the analysis of the rate of change (speed) of up and down ticks separately for a candle’s lifetime in the M1 chart. Whenever we see a high up to down tick speed we have support to go long. Maybe we already can conclude the most probably prize direction by only analysing wicks.
Now I’m really looking forward to hear your opinion on both ideas. Are they practicable and coherent?
 This topic was modified 2 years, 11 months ago by Anti.
Re. 1.
The problem with a ‘representative cycle length’ or ‘dominant cycle length’ (see John F. Ehlers’ publications) is that the concept relies on a high degree of stationarity of your system. Ehlers wrote a lot about that, and as an Electrical Engineer and Radar Specialist he seems to believe in the concept for trading. When using a radar device, the spectrum you’re emitting is controlled by yourself and can be considered stationary, thus the reflected signal you’re analyzing shows very stable conditions.
Trading Intraday FX, there’s no such thing as a ‘stationary’ cycle length (edit: but if you find it, I’m more than interested!). Use the concept to analyze truly seasonal data, like annual crop prices or similar. This will work and certainly support your trading decisions relating to these symbols. In FX, I failed to find price cycles that could be analyzed before the cycle length changes or the market goes flat. What works, to a certain extent, is analyzing intraday volume fluctuations of certain pairs (‘Atlantic’ pairs, not ‘Pacific’ ones).
Keep in mind that you’ll need several full cycles to detect a stable cycle length, if it ever exists. Just analyzing one half cycle or so certainly won’t do.
If you’re looking at ‘true’ cyclic data, try bandpass filters, e.g. Ehlers’ Roofing Filter. A bandpass filter does not require an exact determination of a cycle length, only the estimation of an appropriate upper and lower stop frequency of your band. But it also requires truly cyclic data, and that’s why I don’t believe it will work on FX. Where can we find seasonality in FX prices?
Re. 2.
I’m also considering to test intrabar tick count as an additional signal. At the moment, it’s not my primary focus. It can easily be coded, but can’t be backtested.
s.
 This reply was modified 2 years, 11 months ago by simplex. Reason: clarification
A good trader is a realist who wants to grab a chunk from the body of a trend, leaving top and bottomfishing to people on an ego trip. (Dr. Alexander Elder)
Hi @simplex,
Re. 1. The problem with a ‘representative cycle length’ or ‘dominant cycle length’ (see John F. Ehlers’ publications) is that the concept relies on a high degree of stationarity of your system.
I’m not sure if you can compare my initial thoughts with Ehlers theory (I don’t know enough about it). However, the above attempt does not assume constant cycle lengths to forecast market turning points. It is just designed to measure the recent position within a cycle. The optimal cycle length (equivalent to length of flattest MA) can change from one candle to another. Thus, the length of one cycle to another is probably different, e.g. due to blurring of prize (which revolves around the “true” market value). However, its application and the use of obtained “best recent cycle length” may help to filter some bad indicator signals. Assume that we calculate the stochastic for each “best MA length” for each candle. Then I’d expect to see the stochastic value much less haning above the 80 value during up trends or below the 20 value during down trends as we can see it for smaller settings.
Nevertheless, I can be totally wrong. The idea was just an inspiration. I haven’t performed any tests.
If you’re interested, here are the links to threads that innovated me: here and there.
Re. 2. I’m also considering to test intrabar tick count as an additional signal. At the moment, it’s not my primary focus. It can easily be coded, but can’t be backtested. s.
… one old MT4 problem …
He mentioned that we can use MAs for instance to determine optimal settings for the stochastic indicator. Optimal settings may have reached when the slope of a MA becomes flat. Thus, the most representative cycle length for a particular time point (candle) can be identified by minimizing the sum of squared slope differences.
Least sqares, ok! I’ve had the same idea about three years ago when studying some of Ehlers’ papers, and abandoned it before starting to code.
How many candles is the width of your observation window supposed to be?
A good trader is a realist who wants to grab a chunk from the body of a trend, leaving top and bottomfishing to people on an ego trip. (Dr. Alexander Elder)
How many candles is the width of your observation window supposed to be?
I think it will depend on TF. For M1 maybe something between 500 and 5000 candles. But I really have no idea. Just want to hear your opinion before starting a new journey …
Regarding 2nd idea…
Perhaps this is something you would be interested in:
Hi all. I was also thinking about how to use the settings of MA and stochastic for more savings. Actually I can say that the market has cycles, I have long studied “fractal market analysis” and can visually identify these cycles. But the problem is that the cycles are not periodic and the trade in them is maintained at certain levels, with a statistically precalculated probability. That is, the cycle can at any moment be cancelled, and besides, there are several basic types of loops which differ from each other by a distance in price and time.Regarding the second theme is really interesting and I have a EA which trades based on speed of price movement.
TM
Ничто не ново под луной:
Что было, то и будет.
Я завершаю круг и  вновь
Готов бежать по кругу...Actually I can say that the market has cycles, I have long studied “fractal market analysis” and can visually identify these cycles.
Would you please share a picture on that topic?!
But the problem is that the cycles are not periodic and the trade in them is maintained at certain levels, with a statistically precalculated probability. That is, the cycle can at any moment be cancelled, and besides, there are several basic types of loops which differ from each other by a distance in price and time.
But I don’t think that this would be a problem as we don’t predict the cycle length but calculate for each candle the best fitting (flattest) MA. The only disadvantage in that case would be that we may miss (many) trading opportunities. However, the signals obtained by calculating an Stochastic with those variable cycle length could be great (I’d expect to see in most cases the stoch value only dipping into the overbought/oversold region before returning back).
Regarding the second theme is really interesting and I have a EA which trades based on speed of price movement
I’d like to know more on your general framework. Do you use complete up/down tick speed or only the speed within the wicks? Do you postprocess those tick data? Do you just compare both tick speeds and obtain a signal whenever it’s ratio is above/below a threshold?
Actually I can say that the market has cycles, I have long studied “fractal market analysis” and can visually identify these cycles.
Would you please share a picture on that topic?!
But the problem is that the cycles are not periodic and the trade in them is maintained at certain levels, with a statistically precalculated probability. That is, the cycle can at any moment be cancelled, and besides, there are several basic types of loops which differ from each other by a distance in price and time.
But I don’t think that this would be a problem as we don’t predict the cycle length but calculate for each candle the best fitting (flattest) MA. The only disadvantage in that case would be that we may miss (many) trading opportunities. However, the signals obtained by calculating an Stochastic with those variable cycle length could be great (I’d expect to see in most cases the stoch value only dipping into the overbought/oversold region before returning back).
Regarding the second theme is really interesting and I have a EA which trades based on speed of price movement
I’d like to know more on your general framework. Do you use complete up/down tick speed or only the speed within the wicks? Do you postprocess those tick data? Do you just compare both tick speeds and obtain a signal whenever it’s ratio is above/below a threshold?
Sure.
this is the markup of the fractal analysis. now we can observe an upward cycle of 10 March. A basic move is indicated by white arrows, red – correction. the red line at the level of “media” this is the strongest support if the price goes over it and it fixed 90% probability that the upward cycle will be cancelled. Orange rectangle is the area of the noise , the price can hang out and it will be a correction, but if the price could pass it there will be a distinct impetus to the level of 261.8. At the moment my first target is 161.8 price already took it and since after it has been rolled back to the level of the probability SHUM cancel the upward cycle now 40% and continue upward movement 60%. And based on the characteristics of the pair EU it often overrides the potential, i.e. we are likely to see a pullback or reversal that would mean education nishadha cycle. in advance at the beginning of the cycle it is impossible to say to what level the price will go, so in the fractal analysis makes the prediction for 12 days ahead.Attachments:
You must be logged in to view attached files.Ничто не ново под луной:
Что было, то и будет.
Я завершаю круг и  вновь
Готов бежать по кругу...Regarding the second theme is really interesting and I have a EA which trades based on speed of price movement
I’d like to know more on your general framework. Do you use complete up/down tick speed or only the speed within the wicks? Do you postprocess those tick data? Do you just compare both tick speeds and obtain a signal whenever it’s ratio is above/below a threshold?
I downloaded it a few days ago and have not had time to understand it. but I can share if you are interested.TMAttachments:
You must be logged in to view attached files.Ничто не ново под луной:
Что было, то и будет.
Я завершаю круг и  вновь
Готов бежать по кругу...Regarding 2) you also could have a look at http://www.forexfactory.com/showthread.php?t=536931
Just another, maybe stupid question. Yesterday I’ve tried to filter some wrong signals. My intention was to just use common technical tools (e.g. MAs, Stochastic, …). The problem I was confronted with is that all those “classical indicators” don’t really differ. For some of those indicators this is obvious. For instance, the 50 % line of X stochastic with high/low settings is exactly the midline of a X donchian channel. However, I also tried to use ATR and standard deviations which delivered nearly the same signals as the MA strategy I set up. Thus, the big question is which possibilities remain to filter wrong signals. Any recommendations?
Certainly not stupid!
IMO, the main problem in ‘classical’ TA is that most of those socalled ‘classical’ indicators do more or less the same: some kind of simplified calculus based on slightly different math in hindsight. I think it was Albert Einstein who stated that it’s one indicator for madness to expect different results while always using the same tools.
You can only expect fundamentally different results if you’re applying fundamentally different math, or if you find truly independent data to include in your analysis.
I think all you can do to enhance the success rate of entries based on those indicators is to apply volume as an additional source of information besides OHLC, to analyze your data in different timeframes and somehow bring those analysis to congruence, and to analyze base and quote currency of a symbol separately, i.e. perform a currency strength analysis. Any of these techniques will provide individual strengths and weaknesses.
In MT4, we do not have more information, only OHLC + V. If you want to overcome this lack of information, you have to analyze additional data outside of MT4 (or somehow feed additional data into MT4) or maybe go for manual fundamental analysis and add manual filters accordingly to your MT4 system. Looking at correlated or anticorrelated markets could provide independent information, or adding seasonal information, if applicable.
And finally: what would you expect? If those classical indicators would provide totally contradictory results, you would have a real reason to complain, right?
If you’re getting ‘wrong signals’, it would certainly be interesting to analyze the reasons for this in detail: Is volatility too high or too low? Insignificant Volume? News events? Correlation between currencies of a pair too high? Is one of your indicators just buggy? Is the trend analysis ok, but the entry timing poor? …
s.
A good trader is a realist who wants to grab a chunk from the body of a trend, leaving top and bottomfishing to people on an ego trip. (Dr. Alexander Elder)
Yes, Einstein was a trader …
As an anthropologist/biologist who believes in evolution, I wonder why that many similar indicators have been developed and why they survived/have not been selected. Do they really reveal/represent market behavior?
As a biologist, suppose you’re up to describe the relationship between 8 species who are sharing one common ecosystem (‘the riverside’) in a quantitative way, let’s say storks, frogs, toads, pikes, carps, and some insects. Which data would be at hand and which would you use for a forecast of population cycles of each one of those species?
[Fill in your answer here! I’m confident you could fill in a lot and then have the luxury problem of selection in order to simplify your mathematical model.]
As a trader, suppose you’re up to describe the relationship between 8 currencies who are sharing one common ecosystem (‘the market’) in a quantitative way, let’s say EUR, CHF, GBP, USD, CAD, AUD, NZD, and JPY. Which data would be at hand and which would you use for a forecast of price cycles of each one of those species?
[Fill in your answer here! For technical analysis, we don’t have that much data at hand, certainly there won’t be such a luxury problem.]
For my personal conclusion, you might refer to my previous post.
Looking at one OHLCV timeseries, what usually follows is the eternal hunt for valid ‘signals’ by weighing between integration and differentiation, resulting in smoothness vs. responsiveness.
Looking at several timeseries and extracting currency strength data out of these, we collect data about each currency, thus enabling ourselves to base the analysis of each single pair involved on two individual currency based timeseries, broadening our statistical basis from 1 dimension (1 price) to 2 dimensions (individual strength of base and quote currency).
So by using CS, we are able to extract truly additional information by weighing several timeseries against each other as opposed to studying only one timeseries at a time.
The shere lack of data at hand makes it highly interesting to find additional sources of information.
And your last question: I suggest you feed in price data alongside with RSI, Stoch, CCI, etc. readings in an R or Excel model and run a correlation analysis. Depending on parameters and time windows chosen, the correlation coefficients will spread in a wide range. That’s great, because it means everybody who wants to find a positive correlation visually in order to confirm his / her beliefs, will be able to find such events easily. And it’s dangerous, if you miss to base your statistics on a wide variety of cases to get your feet back on solid ground. So my answer would be: sometimes they do represent prices, but most of the time they don’t.
s.
A good trader is a realist who wants to grab a chunk from the body of a trend, leaving top and bottomfishing to people on an ego trip. (Dr. Alexander Elder)
A great post, @simplex. Really like it!
[ pipatronic ]May 26, 2016 at 7:15 pm #12607Thanks!
And no – this is not my stomach!
s.
Attachments:
You must be logged in to view attached files.A good trader is a realist who wants to grab a chunk from the body of a trend, leaving top and bottomfishing to people on an ego trip. (Dr. Alexander Elder)
How do you know? We’ve never met, have we?
A good trader is a realist who wants to grab a chunk from the body of a trend, leaving top and bottomfishing to people on an ego trip. (Dr. Alexander Elder)
@anti: I remembered our discussion (2 years ago) when I stumbled upon a more recent interesting article: Naive Bayes classifier for signals of a set of indicators, which is also available in German.
Cheers, s.
A good trader is a realist who wants to grab a chunk from the body of a trend, leaving top and bottomfishing to people on an ego trip. (Dr. Alexander Elder)

AuthorPosts
You must be logged in to reply to this topic.