Trading with Market Statistics VI. Scaling In and Risk Tolerance

[jperl]

Talking about burning out neurons, let me see if I can tackle this one from Dogpile:

	Quote:	Dogpile
	the premise of this strategy is that for any non-normal (skewed) distribution, standard deviation will understate the 'true' volatility.

Sorry Dogpile, but this is just plain wrong. The standard deviation computation IS the true volatility, for the distribution in question, no matter what the shape of it, normal or otherwise.

	Quote:	Dogpile
	thus, 2 std devs is not as far away as you might first think by the std dev calculation (ie, 2 std devs means something different if the distribution were instead 'normal' -- normally distributed).

No, 2 std devs doesn't mean anything different. It's just that most people exposed to statistics, have only seen the normal distribution and think the whole world revolves around that (I guess that's why its called "normal"), so they think that 2 SD should always represents 95% of the data for any distribution. The normal distribution,however, is not the center of the statistical universe. There are more different types of distributions than you can count on both your fingers and toes. Academicians have been trying to pigeon hole the type of distribution that markets follow for years. The closest they've come is something called the Pereto-Levy distribution which I don't want to discuss here since it really doesn't have any practical significance for trading. The point is, no matter what the distribution function looks like, the standard deviation is computed in exactly the same way for all of them. How much of the data this represents will of course be different depending on the distributions shape.

	Quote:	Dogpile
	we have observed many cases where price will not return to VWAP -- despite VWAP serving as the 'mean' for the distribution in the std dev calculation.

If you are thinking here that the VWAP is some psuedo mean then that's not correct. The VWAP IS the exact mean for the volume distribution. For any finite distribution of arbitrary shape, you can calculate the mean in exactly the same way.

	Quote:	Dogpile
	thus, for a skewed distribution -- you can kind of think as 1 SD as the 'new mean' (though that term 'mean' probably isn't right -- just think of it as a new potential 'pivot' since price might very well not return to the VWAP price)

Ok, just don't call it a new mean. I've already given it a name. It's called HUP. We will discuss the meaning of HUP and pivots in a later thread

	Quote:	Dogpile
	so skewed distributions are kind of funky -- you are using VWAP in the calculation as you would the 'mean' -- but it really shouldn't be thought of as the mean.

Wrong on both counts. Skewed distributions are funky to you, because you are not used to them, but they have been around a lot longer than any of us have been on this planet.
And once again, the mean for a finite skewed distribution is well defined.

The VWAP is as mean as they get. lol

[Dogpile]

yes I meant that you shouldn't think that 95% of observations should be within 2 std devs if the distribution is skewed -- I should have said that rather than saying 2 std devs is different in its own right. clearly, I am not up to speed on the properties of skewed distributions so this isn't intuitive to me.

instead of saying:

the premise of this strategy is that for any non-normal (skewed) distribution, standard deviation will understate the 'true' volatility.

is it right to say,

the premise of this strategy is that for any non-normal (skewed) distribution, you would expect far more observations to fall 2 std deviations away from the mean than you would if the distribution were normally distributed.

??

re mean and VWAP --- most people used to something like a moving average would expect regression to the mean. but we have stated that this is not necessarily the case with regard to VWAP. thus in this case, 'mean' doesn't mean the same thing as 'mean' -- as in 'regression to the mean'.... or at least, regression to the mean would not be expected until the skew of the distribution changed.

on first read of your reply, it looked like I really didn't understand this at all. but now when I think about it -- what I said was not technically correct but the trading implications weren't far off -- assuming this post here is correct.

[jperl]

	Quote:	Dogpile
	its hard for me to relate to this 'scaling in' thread given the example in the video. the location of the initial short is just a location I wouldn't ever consider shorting NQ -- and I trade NQ every day. I am not saying this method isn't profitable, just that personally -- I place a premium on having good trade location and use a hard but reasonable stop against that -- then often re-enter if I believe my initial read was correct and my stop just wasn't wide enough. I can see the logic in your approach Jerry but I just don't think that is optimal trade location -- hence my question above.

Well Dogpile, I wish I had traded NQ today instead of ER. There were at least four good long entry points including 1 scale-in point. In all trades, the VWAP was above the PVP, so distribution was skewed to the upside as you can see in each of the charts below.

Here are my NQ charts

#1 Enter long at the VWAP 1933.50, exit at 1st SD 1938.25 profit 4.75 pts



#2 Enter long at 1st SD 1939.75 exit at 2nd SD 1945.75 profit 6 pts



#3 Enter long at 1st SD 1943.75
Scale-in at VWAP 1937.25 giving Break even at 1940.50
exit at 1st SD 1944.00 or higher profit 7.00 pts




#4 Enter at 1st SD 1945.00, exit at 2nd SD 1952.00 profit 7.00 pts.


Total profit for the day 24.75 pts

Pretty good I'd say using just simple statistics.

[Dogpile]

nice. yes well today was quite a different set-up than July 25th (the day of the video). I could list all the ways it was different but I know you don't really care.

I did watch the PVP/VWAP relationship at times today and did factor this in as a slight (long) bias to my thinking -- I did make about 10 pts on NQ today. I should have made more but got sidetracked with thinking that the market was going to have a tougher time than it did with all that previous congestion/resistance from last week.

[darthtrader]

comeon guys, save this stuff for last....so over the head of who is following this so far.
HUPS and market rotation theory next??

[jperl]

	Quote:	darthtrader
	comeon guys, save this stuff for last....so over the head of who is following this so far. HUPS and market rotation theory next??

What stuff are you referring to darth?

Next thread will be on counter trend trades at the PVP. We won't get to details about HUP until thread IX or X.

[cooter]

	Quote:	jperl
	Well Dogpile, I wish I had traded NQ today instead of ER. There were at least four good long entry points including 1 scale-in point. In all trades, the VWAP was above the PVP, so distribution was skewed to the upside as you can see in each of the charts below. Pretty good I'd say using just simple statistics.

Jerry,
In your assessment of VWAP and its SD, does price ever reach the 3rd SD level? If not, why not?

[jperl]

	Quote:	cooter
	Jerry, In your assessment of VWAP and its SD, does price ever reach the 3rd SD level? If not, why not?

good question cooter. yes price will reach the 3rd SD at times. This usually occurs on a fast momentum break out. In the last week there has been a lot of this. Extremely difficult to trade, if you didn't catch the move at the start. You don't want to enter a trade when price hits the 3rd SD. A rebound could catch you off guard.