What do you consider noise?
- Thread starter mcteague
- Start date
According to wikipedia:
"Noise is an unwanted perturbation to a wanted signal."
That definition works for me. People who say noise is nothing or noise is everything are talking nonsense IMO. The word has a long-standing meaning, so stop perverting it for your own narrow ends.
Price = Signal + Noise
Signal is that portion of price which is tradeable; noise is that part which is untradeable. The two are not the same and figuring out how to tell them apart is the key to successful trading.
Here's another hint:
Price = Trend + X
If the above equation is true (you must decide), what is best represented by X?
Good luck.
I'm glad you find them interesting, but I also think they're relevant. This is because of the increasing volume of data I gathered for myself regarding indicator performance (if you want to use the silly word "prediction"). It seems to me, so far at least, that the only thing indicators can really predict with some kind of certainty are daily and monthly trends. There have been books written on the subject and papers published too, some more interesting than others. I think there are more people agreeing that noise only somewhat diminishes in those higher time frames.
Also, looking at some of the strategies developed by others, to me it seems (my opinion) that they all somehow amount to an alternative to mean-reversion strategies. Those can work with random entries too, but some people use intraday indicators to make them think their entries aren't really random - at the fundamental level.
So you'd say that the signal is a 20-day moving average? Or what is it?
All we see is "the price", you can make out the signal only in hind-sight and then only partially, or if you had perfect information about the market participants like i already mentioned in my last post.
Noise is tradeable, i'm pretty sure.
Good grief! Why would I say something braindead like that? First of all, contrary to naive trader belief, a moving average is not an indicator; it is a crude lowpass filter, nothing more. A MA crossover might be an indicator (a signal) for those amenable to low-grade trend-following but that's not me.
Past and present data are all you need if you know how to analyze it properly. But don't expect anybody (including me) to just hand you knowledge that people sweated bullets to get.
You're absolutely wrong, I'm sure. But whatever.
What qualifies as noise is largely irrelevant to success. The key is to isolate those data points, patterns, structures etc. that produce a positive expectation for you and then exploit that to turn that expectation into consistent performance within a risk parameter that matches your capital.
What you end up using will be a mere grain of sand compared to the millions of tons of sand available to you. Do not fall into the trap that all or even any of the rest is noise. Remember there are guys making money using lots of the rest of it that could care less about what you value -- rightly value -- because it works for you. When I was a kid I threw side arm and a coach was telling me how to change it. A smarter coach looked at him and said "doesn't it occur to you that he's delivering a pretty good share of strike outs and we are winning with him on the mound".
Noise, like beauty, is largely in the eye of the beholder.
What you end up using will be a mere grain of sand compared to the millions of tons of sand available to you. Do not fall into the trap that all or even any of the rest is noise. Remember there are guys making money using lots of the rest of it that could care less about what you value -- rightly value -- because it works for you. When I was a kid I threw side arm and a coach was telling me how to change it. A smarter coach looked at him and said "doesn't it occur to you that he's delivering a pretty good share of strike outs and we are winning with him on the mound".
Noise, like beauty, is largely in the eye of the beholder.
I actually agree on most of what you said, I guess you misunderstood my tone.
Except for the part where you say I'm wrong.
In theory, with sufficient capital, it's very possible to extract profits from random prices. In fact, just recently I saw G-Bot on this very forum. It auto-develops strategies on random price curves (geometric brownian motion, coin toss, etc) and gives quite a number of positive statistics on a number of such strategies. I looked into it and it's a legit approach.
Here's a link with some real-money results: http://www.elitetrader.com/vb/showthread.php?s=&threadid=222126
By creating random corridors you can extract profits from prices moving through them, and loosing when it changes a corridor. Makes sense if you think about it, it's one approach to mean-reversion in effect.
Nobody said anything, thankfully. I was just saying I had my reasons for saying that about indicators (and how they don't work in noise, thus confirming the effect) in case it was misunderstood, and since we agree, it would be interesting to also get an idea what path you took to reach the same conclusion. I have no idea how many people do their own statistics or just read books on the subject, or why they're saying what they're saying, so i'm just curious...
