I just wanted to say the best system doesn't work, if You don't stick to it. That's the psychological part of the game. But You are right. If I had a system, that worked 90% of the time without too much drawdowns, then one would not have to make decisions at each trade. And decisions are part of Yourself, they are not part of the system. You have to decide every minute - do I trade now or not. If there would be a system that worked in 90% of the time correctly, then sooner or later everyone would know this system and trade AGAINST the system's rules to make a gain. It 's like chess. Imagine You would know what Your opponent is going to do on the next move, then You would for sure make ANOTHER move just to destroy his planned strategy and maybe implant Your own strategy into the game. In the market, when everbody knew - oh, now is the time to buy XYZ, because perfect system ABC is telling me that - then for sure some smart guy would buy a little earlier and another guy again a little earlier, and so on, until Your 'perfect' system is not perfect any more. Every perfect system is doomed to die.
I did a study several years ago on Hurst factors for different types of markets. For most future markets the factor ranged from .52 to .56, great for trend followers. The SP500 had a Hurst fator of .47, not so good for trend following. jj
Why is 0.56 good for trend following? What is the confidence level for futures? How did you establish the benchmark for what is trend following and what is not?
The Hurst Factor is calculated by taking the Std Dev of the price changes over different time periods and then doing a data regression using these std devs along with the time element. A .5 is random. Anything higher or lower would suggest that the past price changes have an influience on the future changes. Over .5 would suggest that the longer term price changes would be larger than expected in relation to the shorter ones. JJ
I think what JS11374 and I are asking is the confidence interval on the Hurst exponent calculation. What is the chance that a random price series might have a Hurst exponent of 0.56 or higher? Without either an analytical calculation of this probability or a Monte Carlo simulation of a random market, it is very dangerous to assume that a Hurst exponent 0.56 means that the market is profitably tradable using trend following techniques. Moreover, one needs to understand how the Hurst exponent varies over time. Although the value may average 0.56 over a multi-year timeframe, it could be below 0.5 for significant periods. If the value is truly below 0.5 for an entire year (but masked by the extremely high values in other years), then trend following traders might endure a 1-year losing streak (YIPES!). The point is, NEVER go by the averages of anything. If one does not understand the possible statistical variations that can corrupt the measured "average" value, one runs severe risks in trading according to those averages. -Traden4Alpha
True. The Hurst Factor is only good for looking at the past performance and will not predict the future. However, if I wanted to trend follow I would pick a market with a high Hurst. If I wanted to counter trend I would look for a low Hurst. JJ
JS, In my studies I found that the Hurst factor will generally rbe in the .4 to .6 range. A .6 would be very high and a .4 very low. JJ
JJ, is the 0.4-0.6 range across various instruments or different time frame for a single instrument (intraday, 5-min, daily, weekly, etc...). The reason why I ask is because when I ran such tests in the past, it seems that it's more or less possible to establish non-brownian behavior on just about anything, given enough undersampling. For example, I generated ten million random number. The hurst is about 0.501 for the series. Good enough. Then I picked a continuous subset of about 10,000 numbers. The hurst exponent is about 0.72. Repeat this with different subsets about 100,000 times, I got Hs ranging between 0.2 to 0.85. So my conclusion is, unless you are confident that you have the entire global set for the stock prices (which, technically reads: all past and all future), this test, like most others, is purely academic.
Thank you JS11374!!! Finally someone who has actually looked at the statistical distribution of Hurst values. Do you recall the standard deviation on your 100,000 runs of 10,000 samples each? (I'm guessing it was about 0.07??) -Traden4Alpha