Moving Average or Linear Regression?

P

programtrader

Does anybody has experience working with these smoothing techniques, that could point any advantages or disadvantages of using MA's or LR's for:

a) trend direction measurement

b) mean reversion measurement

LR's for the same period seem to be more prone to whipsaws, ence better for the measurement of mean reversion patterns?
 
Quote from jasonbraswell:

How is it leading, Maestro?

I don't know too much about splines, except that they're apparently 4th degree polynomials that fit data?
More...
No, you can use cubic (3rd degree) splines.

I assume MAESTRO meant cubic spline smoother, because cubic spline interpolation fits every point, and so it models noise right along with signal. :eek:

But there's a problem even with the smoother: the endpoints are fitted, so you don't get smoothing on the most recent point, which is where you (a trader, not an econometrician) need it the most, usually. :(
 
I prefer a first order infinite impulse response filter myself (better known as an alpha filter to the unwashed), but then again I am infinitely impulsive. However, I cheerfully defer to the artist formerly known as Abogdan. He is the smartest mathematician here. I had a girlfriend once who had a cubic spline. She fit very nicely.
 
Quote from MAESTRO:

Cubic Spline Interpolation is a leading predictor
More...

I once tried to do this with company earnings/sales, and because it was non-linear, I got weird numbers, for example, it would go

1,2,3,4--->6*10^4000

Or something like that. It fit the existing data points fine but extrapolating was almost impossible. In the end, I just sucked it up and went back to y=mx+c
 
Quote from programtrader:

Does anybody has experience working with these smoothing techniques, that could point any advantages or disadvantages of using MA's or LR's for:

a) trend direction measurement

b) mean reversion measurement

LR's for the same period seem to be more prone to whipsaws, ence better for the measurement of mean reversion patterns?
More...


1. LR's slope is a lagging indicator
2. MA by itself is also a lagging indicator

Thus, using #1 or #2 your trend identificatoin will be lagging.

For mean reversion, the more important element is breach of extremes - you've come up with 2 mid points, what about the offset from the midpoints?

In my opinion thats more important.

Take a look at a scalping system I posted,
http://newsletter.neoticker.com/?p=53

The setup is simple but it continue to work year after year. And the interesting point is, it depends on the limited time a trend has on its continuation at the extremes.
 
Quote from Albert Cibiades:

I prefer a first order infinite impulse response filter myself (better known as an alpha filter to the unwashed), but then again I am infinitely impulsive. However, I cheerfully defer to the artist formerly known as Abogdan. He is the smartest mathematician here. I had a girlfriend once who had a cubic spline. She fit very nicely.
More...
"Do not buy any moving average based system!" -- abogdan

:D :D :D
 
You must log in or register to reply here.
Back
Top