Can anyone code John Ehler's Empirical Mode Decomposition
-
Hi, Is anyone able to covert this Empirical Mode Decomposition easy language code to PRT? Thanks in advance.
it’s a “novel approach for cycle and trend mode detection.” https://www.mesasoftware.com/papers/EmpiricalModeDecomposition.pdf
BandPass Filter Code:
Inputs: Price((H+L)/2), Period(20), delta(.1); Vars: gamma(0), alpha(0), beta(0), BP(0); beta = Cosine(360 / Period); gamma = 1 / Cosine(720*delta / Period); alpha = gamma - SquareRoot(gamma*gamma - 1); BP = .5*(1 - alpha)*(Price - Price[2]) + beta*(1 + alpha)*BP[1] - alpha*BP[2]; Plot1(BP); Plot2(0);Code to Extract the Trend:
Inputs: Price((H+L)/2), Period(20), Delta(.1); Vars: gamma(0), alpha(0), beta(0), BP(0), Trend(0); beta = Cosine(360 / Period); gamma = 1 / Cosine(720*delta / Period); alpha = gamma - SquareRoot(gamma*gamma - 1); BP = .5*(1 - alpha)*(Price - Price[2]) + beta*(1 + alpha)*BP[1] - alpha*BP[2]; Trend = Average(BP, 2*Period); Plot1(Trend); Plot2(0);Code for the Empirical Mode Decomposition:Inputs: Price((H+L)/2), Period(20), delta(.5), Fraction(.1); Vars: alpha(0), beta(0), gamma(0), //delta(.4), BP(0), I(0), Mean(0), Peak(0), Valley(0), AvgPeak(0), AvgValley(0); beta = Cosine(360 / Period); gamma = 1 / Cosine(720*delta / Period); alpha = gamma - SquareRoot(gamma*gamma - 1); BP = .5*(1 - alpha)*(Price - Price[2]) + beta*(1 + alpha)*BP[1] - alpha*BP[2]; Mean = Average(BP, 2*Period); Peak = Peak[1]; Valley = Valley[1]; If BP[1] > BP and BP[1] > BP[2] Then Peak = BP[1]; If BP[1] < BP and BP[1] < BP[2] Then Valley = BP[1]; AvgPeak = Average(Peak, 50); AvgValley = Average(Valley, 50); Plot1(Mean); Plot2(Fraction*AvgPeak); Plot6(Fraction*AvgValley);Hi Nicolas, if you manage to get the time this could be worth adding to the Ehler’s indicators? I don’t know how it’ll compare to the Bandpass Filter?
Also re: the Universal Supersmoother Oscillator, I have an Ehler’s MESA Period indicator to get cycle period lengths eg for a 1 hour or 10 min chart for example. Is that value suitable to enter as the frequency for the Bandedge in the Univ. Osc.? When it is used (usually values around 20 to 30) it never performs as well on backtests as say a 100 or 150 bandedge setting?
Often I see very good results on the 10 min chart and 4 hours for the Univ. Osc. but not on the hour chart despite changing the bandedge and the 20/80% cut off values?
Also when you get good results with the Bandedge/cut off values (without doing a full optimisation, i.e. just trying various settings e.g. change Bandedge from 100 to 150 and filter from 0.8 to 0.9), you can get the system to work over certain random periods of time like a month from 3 total months or for a week (using a 10 min chart) or even over years (using a day chart) but not 0ver the entire period that’s available to test (not ideal as you ”burn” the data in the sense that it’s now optimised). I can randomly pick many single week long periods at a time over a three month period (10 min chart) and they will be profitable 9 x’s out of 10 but just not over the entire period. I was hoping this Empirical Mode Decomposition indicator could be used in conjunction with the Univ. Osc.
Also when I did a full optimisation (by adding the indicator code to the system code) for bandedge 5 to 150 in steps of 5, it returned the same maximum gain (20.69% pls see screen) for every different bandedge value?
Is the Walk Forward analysis the only way to go to test this phenomena where it’s profitable in many random sample periods but not over the entire period? Shouldn’t we just optimise to the latest current market conditions anyway?
p.s. I cannot see the post I made in the forums two or three days ago about the Clenlow ATR Plunger? Subject Heading: “This looks interesting..“ and it has 4 charts. It’s not in the “my topics” in my account on here?
Be interested to hear your thoughts,
Cheers
Brad
Please find below the empirical mode decomposition indicator code:
//PRC_EmpiricalModeDecomposition | indicator //20.02.2017 //Nicolas @ www.prorealcode.com //Sharing ProRealTime knowledge //converted from EasyLanguage code Price = medianprice Period=20 delta=.1 Fraction=.25 if barindex>Period then beta = Cos(360 / Period) gamma = 1 / Cos(720*delta / Period) alpha = gamma - SqRt(gamma*gamma - 1) BP = .5*(1 - alpha)*(Price - Price[2]) + beta*(1 + alpha)*BP[1] - alpha*BP[2] Mean = Average[Period*2](BP) Peak = Peak[1] Valley = Valley[1] If BP[1] > BP and BP[1] > BP[2] Then Peak = BP[1] endif If BP[1] < BP and BP[1] < BP[2] Then Valley = BP[1] endif AvgPeak = Average[50](Peak) AvgValley = Average[50](Valley) endif return Mean as "mean", Fraction*AvgPeak as "avg peak", Fraction*AvgValley as "avg valley"If it’s ok for you, I’ll add it into the library for everyone’s benefit.
- You must be logged in to reply to this topic.
Summary
This topic contains 9 replies,
has 3 voices, and was last updated by 
8 years, 11 months ago.
Topic Details
| Forum: | ProOrder: Automated Strategies & Backtesting |
| Language: | English |
| Started: | 01/24/2017 |
| Status: | Active |
| Attachments: | No files |
