Quants are afraid of Mark Jurik because he writes books and sells add-ins for Tradestation. They also fear him because of his large and growing portfolio of US Patents. Have a look at the list, which you can find at the US Patent and Trademark Office website: http://bit.ly/d5WKOn
"Searching US Patents Text Collection... Results of Search in US Patents Text Collection db for: IN/"Jurik, Mark": 0 patents."
I don't think you can patent an algorithm, but even if you could, why would you reveal said algorithm in a patent knowing every programming-savvy trader in the world would just co-opt it illegally to keep from paying you for it? Best to keep it a trade secret with encrypted software.
You can't have it both ways. Either it outperforms the JMA or it doesn't. Fact: the HMA underperforms the JMA. Next.
What metric were you using when you said this: "I have personally seen reverse engineered JMA , and RSX that are identical to the real thing and they were mediocre." I can't do RSX but I have a custom ama that is superior to the KAMA and it isn't mediocre. And it isn't the JMA either. No PHASE parameter, for one thing.
That was another poster's quote, but I ask because it's hard to really say what's better or not, without some type of metric for comparison. As another poster pointed out, what good is a diagram of the response to a perfect edged step input? It is meaningless and arbitrary with regards to trading. I hope you don't rely on that diagram as the arbiter of superiority over filters, because it doesn't really mean anything by itself, other than one response has less overshoot... the obvious trade-off is that it also responds slower, which may or may not be good depending on the data it's filtering. You could pick out several filters; Hull, Jurik, KAMA, etc that perform well based upon a small snippet of market data, which is cherry picking and a general problem with TA to begin with (one of the reasons you don't see quants discussing it much). In order to properly compare them or ascribe any usefulness, there must be either a metric or system to demonstrate their usefulness towards some specific trading application. And the fact that they rely on parameters means that you need to validate them out of sample at a minimum, which you won't see on his website. Secondly, if it is proprietary and you don't have control over the parameters, then you are pretty much at the mercy of his in sample optimized values, which don't necessarily translate well OOS.
Sorry for the mix-up, dtrader98. You're absolutely right, an objective trade-free metric is needed. I'll give it some thought.
The Jurik smoothing includes 3 stages: 1st stage - preliminary smoothing by adaptive EMA: MA1 = (1-alpha)*Price + alpha*MA1[1]; 2nd stage - one more preliminary smoothing by Kalman filter: Det0 = (Price - MA1)*(1-beta) + beta*Det0[1]; MA2 = MA1 + PR*Det0; 3rd stage - final smoothing by unique Jurik adaptive filter: Det1 = (MA2 - JMA[1]) * (1-alpha)^2 + alpha^2 * Det1[1]; JMA = JMA[1] + Det1; where: - Price - Price Series - alpha - dynamic factor(will be described below) - beta - periodic ratio = 0.45*(Length-1)/(0.45*(Length-1)+2) - PR - Phase Ratio: PR = Phase/100 + 1.5 (if Phase < -100 then PR=0.5, if Phase > 100 then PR=2.5). The Dynamic Factor is periodic factor (beta) raised to a power (pow): alpha = beta ^ Pow, where: - pow = rVolty ^ pow1 - rVolty - relative price volatility - pow1 - power of relative volatility with following formula: pow1 = len1 - 2 (if pow1 < 0.5 then pow1 = 0.5), where len1 - additional periodic factor: len1 = Log(SquareRoot(len))/Log(2.0) + 2 (if len1 < 0 then len1 = 0). Thus you can see that the Dynamic factor is based on the relative price volatility giving the required adaptability for this kind of the price filter. The formula for relative price volatility is rVolty = Volty/AvgVolty (if rVolty > len1^(1/pow1) then rVolty = len1^(1/pow1), if rVolty < 1 then rVolty = 1),