Very well said, and that statement espouses my personal trading philosophy as well. I use range bars in both my manual and automated trading, to consolidate the time component into a range - thus priming it for multivariate analysis.
Hey bluelou, I ran some of the numbers on the link you referenced. Although, his idea is very intuitive and simple (which I like), there is a pretty strong systematic error bias that you should be aware of. 1) He asserts that brownian walk should give D=1.5, H=.5 I ran his formula over many trials of a brownian walk and it averaged only 1.4 (H=.6). This is also the number he arrived at after huge runs of monte carlo. He tries to explain it away, by explaining some difficulties in generating true brownian walk (particularly non infinitesimally small sample steps). However, I ran the same exact series I generated on a true R/S algorithm and I get exactly H=.5 as expected. So you might try to calibrate and or even look at the linearity of the D and H values over a sweep of different types of behaviors (trend, rand, mean rev, etc..) as it may need to be calibrated. The extreme ends (i.e. perfect white, and perfect ramp) are also slightly off; uniform white h= .14 GBM rand h = .6 trend ramp h= .93 I have not yet verified if the intermediate ranges are linear, either. Haven't even looked at forward relationships or actual market series yet. But these systematic biases need to be calibrated and compensated for, if you want to have an accurate numerical reference for regime types. Just something to consider. P.S. I got your point on VHF; interesting, but treading into pseudo understanding of underlying reasons IMO (as you mentioned random vs. non-random).
dtrader98, Thank you for the feedback. VHF is like all of the other TA indicators. A very primitive filter, but it can get the job done. If you're using an off-the-shelf stats/econometrics package such as Matlab would you mind sharing your simulation workfiles for the Hurst work? I should get back into using my econometrics s/w and have never used it for simulations. I have Matlab, EViews, SAS, and can get academic versions of everything else. Thanks again, Lou
Hi, I would argue that the sail boat argument is not appropriate in this case. The problem in trading is that trends occur randomly. Therefore you see them only after the fact, when opportunity is already missed. Every time I developed a trend following system that used some sort of trend signal (predictive model), I got bad results. When I developed a trend following system that didn't look for any signals (anticipating model), did not measure any conditions, the results I got were positive and much better. The point I'm trying to make is that anytime you measure market conditions, you're looking at the past market conditions. If you are trading the past conditions at the present, it's no better than random (at least that's what my experience was).
I'm not trying to play games with you here but the model wasn't strictly trend following if it was also predictive.
bingo. everything you need is right there in pneuma's post. read it carefully. there are really only three market conditions at any given time. learn them, trade them all, and trade small. the frequent trading will compound your account quickly with very little drawdown.
I agree with the last two posts concept. I found it works for me. It's just difficult to do unless you build your own software or (dare I say it) use tickzoom. I'll join a discussion of the different markets. I tend to use 4 major market conditions like so... Trend, CountTrend, Chop, and Sideways But it's easier to even split them into LongTrend, ShortTRend, CounterLongTrend, CounterShortTrend, LongChop, ShortChop, and Sideways. That makes 7. I'm curious and willing to learn how you come up with 9. Perhaps, the transitions between them? Or variations on the sideways. Dunno. Wayne
sounds to me like your trying to create the 'holy grail' system, you think everyone else is wasting their time on.