Zactly.
Shortcut to this thread is to do the back testing.
Then easily decide what to ignore.
Low Pass vs SMA
- Thread starter luckyputanski
- Start date
Shortcut to this thread is to do the back testing.
Then easily decide what to ignore.
I backtested a bunch of MAs, and I liked Ehlers work the most, especially the instant trandline indicator of his because it made so much sense on paper. In practice, they didn't show much value thus far, but i'm still collecting data. On the other hand, it looks like they're useful for a special case such as mean reversion when doing pairs trading (or monitoring a large number of instruments and deriving from there), but only to some extent.
As for low pass vs no low pass, in terms of math it's almost the same thing so you won't get much difference. Depends on how many backtests you do though (100 or 1 billion?) and how you construct your systems (entry/exit types, conditionals, etc). You might run into a batch of system constructs that favour MAs, but I wouldn't base any conclusions too early if I were you. Just my 2c...
As for low pass vs no low pass, in terms of math it's almost the same thing so you won't get much difference. Depends on how many backtests you do though (100 or 1 billion?) and how you construct your systems (entry/exit types, conditionals, etc). You might run into a batch of system constructs that favour MAs, but I wouldn't base any conclusions too early if I were you. Just my 2c...
You might find that if your combination is mean reverting, then all oscillators start to work. Now .. is your combination really mean reverting?
Finally had a chance to get back to this and code it.
Very interesting!
And thank you very much!!!
Most of the time, some interesting differences, but no net benefit to signals. BUT, sometimes it shows a top or bottom much much sooner. In one example against one second data, where EMA lags the EMA Fast Slow Cross by 409 seconds, the Fast as a Zero Lag EMA beats the cross by 5 seconds, and the Fast Slow Cross of Zero Lag EMAs beats it by 13 seconds.
By playing with periods, building the Fast Slow Cross with Zero Lag EMAs got 26, 33 and even 79 seconds sooner; but that's curve fitting for these results.
Will take more testing, then backtesting to see if this will make a meaningful change to signals, but very intriguing.
And I'll have to code the others to see how they compare...
No joy.
Often gave in-earlier and out-earlier for a a greater percentage of the trend, BUT the increased noise meant that it would often be thrown out then back into the trend during it's run, so trading fees eat up the extra percentage of the trend, and that's assuming something that's liquid enough to avoid slippage. I can work around the noise if I:
- curve-fit to each trend, but cherry picking after the fact isn't possible live,
- smooth it, but then I loose the improvement in lag,
- add additional criteria for staying in the trend, but that costs me on the exit, losing the exit advantage and costing me the gain from the early entry.
So the consequences of the extra noise that comes with the lowered lag means I can't find a way to make a net advantage.
Sometimes it will work better than SMA.
Sometimes it will work worse than SMA.
It depends on how in tune price movement is with the settings of the MA.
There is no way to know this ahead of time.
Sometimes a 10 period SMA is perfect for catching trends. Sometimes a 10 period will get you stopped out every single time but a 27 period SMA would've been profitable. Or 53. Or 62. Or any other setting that happens to work well with what price did for a given time period.
"Smoothness" is relative to the period being used.
HMAs and JMAs are extremely smooth but work no better than any other MA, that is, sometimes they work perfectly and other times they don't work at all.
Sometimes it will work worse than SMA.
It depends on how in tune price movement is with the settings of the MA.
There is no way to know this ahead of time.
Sometimes a 10 period SMA is perfect for catching trends. Sometimes a 10 period will get you stopped out every single time but a 27 period SMA would've been profitable. Or 53. Or 62. Or any other setting that happens to work well with what price did for a given time period.
"Smoothness" is relative to the period being used.
HMAs and JMAs are extremely smooth but work no better than any other MA, that is, sometimes they work perfectly and other times they don't work at all.
Have a read and listen to some of Ehlers latest work on spectral dilation.
http://www.stockspotter.com/In/TechnicalPresentations.aspx
He doesn't even recommend use of a MA any more, but instead advises to use the super smoother. He is also supposed to have a new book coming out in the Fall.
http://www.stockspotter.com/In/TechnicalPresentations.aspx
He doesn't even recommend use of a MA any more, but instead advises to use the super smoother. He is also supposed to have a new book coming out in the Fall.
Thank you!
I implemented Super Smoother with the java sqrt(2), Math.E and Math.PI instead of constants, just because.
Sure follows the price when smoothing the price. Having never seen anything follow the price so well, I don't know how to make use of this.
For my one second data, it works very well in filtering out what I call micro noise, but leaves the longer noise I'm looking to filter. Extending the period away from SS's optimized-for-10-bars smooths out more noise, but introduces too much lag for it to replace any of my MA indicators. Unless I can figure out how to tune it to longer periods - I'm too new to digital filtering other than averages to know where to start.
But I'll test for using SS where I'm currently smoothing indicators with MAs, where the 10 bar tuning should be just about perfect.
And then there's all the other filters coming from Ehlers to explore.
Thanks again,