Trading Strategies Based On Digital Filters - page 29
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Hi,
It's nice to see that the discussion is going on :-)
I was on holidays for 3 weeks and I'll have a look on all what I missed to follow the discussion with you all.
Thanks a lot to everybody!!
I would like to begin a cycle of articles about time series analysis. If anyone is interested, I will continue and complicate the material.
In my opinion spectroanalysis usage for currency quotations without preliminary extracts preparation is incorrect. Taking into account that the research of some quotations shows that we deal with non-stationary time series , and spectroanalysis methods are suitable only for stationary time series we shall make a spectoanalysis of daytime quotations gbpjpy, all extracts and half of extracts as we see from pictures spectrum of quotations “vary”.
pic.1 GBP/JPY D1 full time series
pic.2 GBP/JPY D1 half time series
In mathematic statistics for transition from non-stationary process to stationary different transformations are used. Their essence consists in the following. Let’s assume that there is a row R0 = {R0.1, R0.2, …, R0.n} then differential transformation of 1-st order from R0 will be number R1 = {R1.1, R1.2, …, R1.n-1}, where R1.i = R0.(i+1)-R0.i . As a matter of fact, the sequence received by the following way, is a number of increments of initial process.
Let's make spectroanalysis for the received sequence. Also as well as in an example for regular extracts we shall take half of artificial extracts and all artificial extracts.
pic.3 GBP/JPY D1 artificial time series
The figure shows the following peaks: 106, 63, 48, 38, 33, 27 .....
You can use this peaks for EA development. We may have good results if we put filters on the received frequenecies. My result:
Period Daily (D1) 2000.06.29 - 2008.02.27
Initial deposit 10000.00
Total net profit 106100.66
Maximal drawdown 16.44%
I wanted to liven this thread up a bit....so here is a trade i put on tonight....we shall see how it goes....the 4hr trend is up....the short may be a retracement that lasts another 6 hours or so.....SATL below RSTL and both sloping down....STLM neg.....3 cycles pointing down...
hope you guys are doing well...
clLooking good, dude. Keep us posted as to how your testing goes.
Note = Be careful using MTF indicators. I'm sure you know the reason why I say this so I do not wish to patronize you. The following explanation is for those who may be unaware:
MTF indicators can and will update closed bars on the current timeframe if the signal on the higher timeframe changes before the close of that bar.
Ex:
Lets say you're using an MTF MACD on M15, w/ the parameters set to H1. @ 00:45, the previous 3 bars are blue, signalling a possible rally. You enter a long order and walk away. You return @ 01:00, and the past 4 bars are now red, signalling a possible drop. You start thinking that either your crazy or those bars changed colors AFTER they "closed".
Fact is that they did, indeed, close on M15 but it was still an open bar on H1. It didn't technically "repaint", at least not in the traditional sense anyway. This is a more extreme case, but the further the distance between current time-frame and the time-frame being monitored using the MTF indicator, the greater the chance of such an occurence. This is the drawback to MTF indicators.
What I like to do is Look @ M30 on M15, H1 on M30, and sometimes H4 on H1(though not often). Easier to stay out of trouble.
My idea was do so something like Vladimir Kravchuk, explained here. Might use a faster timeframe thought and just FTLM, FTLM_KG a.k.a. MTLM, and STLM and RBCI all optimized to the pair and timeframe.
Looking good, dude. Keep us posted as to how your testing goes.
Note = Be careful using MTF indicators. I'm sure you know the reason why I say this so I do not wish to patronize you. The following explanation is for those who may be unaware:
MTF indicators can and will update closed bars on the current timeframe if the signal on the higher timeframe changes before the close of that bar.
Ex:
Lets say you're using an MTF MACD on M15, w/ the parameters set to H1. @ 00:45, the previous 3 bars are blue, signalling a possible rally. You enter a long order and walk away. You return @ 01:00, and the past 4 bars are now red, signalling a possible drop. You start thinking that either your crazy or those bars changed colors AFTER they "closed".
Fact is that they did, indeed, close on M15 but it was still an open bar on H1. It didn't technically "repaint", at least not in the traditional sense anyway. This is a more extreme case, but the further the distance between current time-frame and the time-frame being monitored using the MTF indicator, the greater the chance of such an occurence. This is the drawback to MTF indicators.
What I like to do is Look @ M30 on M15, H1 on M30, and sometimes H4 on H1(though not often). Easier to stay out of trouble.
My idea was do so something like Vladimir Kravchuk, explained here. Might use a faster timeframe thought and just FTLM, FTLM_KG a.k.a. MTLM, and STLM and RBCI all optimized to the pair and timeframe.yea the mtf thing definetly has some "issues" in terms of strict interpretation....
i second ffl's comments on robertinno's thread...could you explain further..
cl
Robertinno,
Of course we're interested. The more, the merrier. Thanks for joining us.
?Questions?:
What is stationary time-series and what is non-stationary time-series
Can you provide more detail as to the meaning of "preliminary extracts preparation". I can visually see the difference between the two w/ the screenshots you posted comparing your spectrum results to that of the DFG software spectrum analyzer.
W/ you using a "half series", are you incorporating some of Hurst theories on using "half cycles" or am I way off?
So am correct in understanding you to say that if we were to use this "correct" method of spectroanalysis, we would generate better digital filters?
How do you feel about using "Goertzel algorithm" vs. "Maximum Entrophy method", as mentioned a few pages back, beginning here
For a complete beginner to these concepts, is there any readings you personally would recommend?stationary time-series - time series with constant on time probability characteristic: average of distribution=const and mean square deviation = constant
artificial time series
1-Robertinho..I quote from your previous posts :
"In mathematic statistics for transition from non-stationary process to stationary different transformations are used. Their essence consists in the following. Let’s assume that there is a row R0 = {R0.1, R0.2, …, R0.n} then differential transformation of 1-st order from R0 will be number R1 = {R1.1, R1.2, …, R1.n-1}, where R1.i = R0.(i+1)-R0.i . As a matter of fact, the sequence received by the following way, is a number of increments of initial process.
Let's make spectroanalysis for the received sequence. Also as well as in an example for regular extracts we shall take half of artificial extracts and all artificial extracts. "
If I understand it correctly,you are proposing to run the spectrum analyzer,not on original price series,but on an artificial one that is basically a time series of CHANGES IN PRICE ?Please be so kind to clarify,and ,if possible,explain how you prepare a time series with an example of both the original and the artificial one....even very simple ones might succeed at explaining how to do it.
Thanks.
2-FFL:In Layman terms a stationary time series is a series with repetitive cycles,cycles are always the same,they do not change in time..obviously this is not so for financial markets.The trick is to filter out the non repetitive cycles and work only with the repetitive ones if any...there are several ways to do so(Bartels test,etc),and what Robertinho proposes looks interesting as a practical way to do so.
1-Robertinho..I quote from your previous posts :
If I understand it correctly,you are proposing to run the spectrum analyzer,not on original price series,but on an artificial one that is basically a time series of CHANGES IN PRICE ?Please be so kind to clarify,and ,if possible,explain how you prepare a time series with an example of both the original and the artificial one....even very simple ones might succeed at explaining how to do it.
Thanks.
Correct. I have used time series increment close (i) = close(i+1)-close(i), close(i+2)= close(i+2)-close(i+1) ....... close(n)=close(n+1)-close(n) for spectrum analyzer.
I've messed with it this morning.....i think the "process" has something to do with this...
export the data to excel....make the needed calculations so that you are finding the change between n's.....paste in a new worksheet those values ( paste special = values)....importing back into metatrader is the next step i think.....otherwise, the digital filter softare doesn't seem to recognize a excel .csv file...it says that is one of the file types it recognizes...but i get errors trying to open it directly......
i've gotten a few occurences to show up in the digital filter software...but then the errors come like i said....i did the rate of change for all open, high, low, and close.....as well as just close.....
on the right track i think, but needing further comments...
cl
I have used excel for time series preparation