Trading with Market Statistics XI. HUP

[jperl 11]

I understand that SD quantitatively measures the volatility. But I still wish to know why SD1 and SD2 are hold up prices but not SD1.5 or SD 1.3 or anything in between? Is there some discontinuity in the distribution in the SD1 and SD2 level so that when price goes there it tends to hold up or slow down? Basically, why those particular points?

 

As far as HUP is concerned, there is nothing special about SD1. Every bar on your chart has a high and a low, each of which is a HUP. The question then comes down to which of the HUP's do you wish to trade against. The SD values are useful in that they represent a quantitative measure of minimal market movement. So if you enter a trade at SD1.5 say, your expectation is that the market should move at least one standard deviation

[n123 10]

Every bar on your chart has a high and a low, each of which is a HUP.

 

According to this, and the fact that we can create bars in infinitely many ways (5 seconds bar, 7 seconds bar, or 23.5 seconds bar... etc.) can we say that in fact every price can be a HUP?

 

The question then comes down to which of the HUP's do you wish to trade against. The SD values are useful in that they represent a quantitative measure of minimal market movement. So if you enter a trade at SD1.5 say, your expectation is that the market should move at least one standard deviationa HUP.

 

Then... here can we say that any point from SD1 to SD3 are possible entry points as we can expect price to move at least one SD from there? So SD1.3 SD 1.7 SD 2.4 all work (as good as SD1 or SD2?) ?

 

I may be way off and missing some important points. Please forgive me if I sound rude or challenging, I just wish to understand this fascinating subject (I like maths!). And as English is not my first language, I couldn't express very well.

 

Thank you.

Edited August 16, 2009 by n123

[jperl 11]

According to this, and the fact that we can create bars in infinitely many ways (5 seconds bar, 7 seconds bar, or 23.5 seconds bar... etc.) can we say that in fact every price can be a HUP?

Well, I wouldn't say every price, but I would agree that you will find HUP on every time scale. In fact if you look at a 5sec chart, you will see this to be the case.

 

 

Then... here can we say that any point from SD1 to SD3 are possible entry points as we can expect price to move at least one SD from there? So SD1.3 SD 1.7 SD 2.4 all work?

 

On the time scale that you are trading, your expectation should be that upon entry, what ever price that is, price action should move the market plus or minus one standard deviation. So let's say you enter the market at SD1.3 in your terminology. Your expectation as a minimum should be that price will either move to SD0.3 or SD2.3.

[atto 20]

I'm just as casual observer, as I don't personally trade this way, though I do have a decently strong background in math and statistics. However, I've had a lot of the same thoughts n123 is having regarding this methodology. Thanks for entertaining us

On the time scale that you are trading, your expectation should be that upon entry, what ever price that is, price action should move the market plus or minus one standard deviation.

Why is this? If 1SD has no intrinsic significance besides a measure of volatility -- that is, there's nothing super special about 1SD vs 1.3SD vs 3SD, outside of different degrees of volatility -- why should we "expect" price to move 1SD from entry? There aren't a ton of people trading this way, so it's not a self fulfilling prophecy. Therefore, if the method's valid (which I trust that it is), there has to be some other truth or reason as to why markets move in 1SD increments. Is it just some (unexplainable?) bias markets have?

 

Also, if pretty much any price could be a HUP (which seems to be the case, since we could be entering at 1SD, 1.3SD, 2SD, 4SD, etc), especially when you consider all the timeframes, why would the market respect 1SD movements from there?

[jperl 11]

If 1SD has no intrinsic significance besides a measure of volatility -- that is, there's nothing super special about 1SD vs 1.3SD vs 3SD, outside of different degrees of volatility -- why should we "expect" price to move 1SD from entry?

Atto, I think perhaps you have answered your own question. The computed standard deviation IS the volatility. Moreover, as I believe I pointed out in thread IV, when computed with respect to the VWAP it represents the smallest standard deviation possible for the data.

For example let's say the standard deviation is 10 ticks. This represents the data volatility. When you enter a trade, regardless of what the price is, you should expect the price to move at least 10 ticks. What of course you don't know, is whether it will move up 10 ticks or down 10 ticks and whether it will do so in a linear fashion. It may move up 5 ticks and then move down 10 ticks. The point is you shouldn't expect the market to move say 20 ticks after your entry.

[atto 20]

Atto, I think perhaps you have answered your own question. The computed standard deviation IS the volatility. Moreover, as I believe I pointed out in thread IV, when computed with respect to the VWAP it represents the smallest standard deviation possible for the data.

For example let's say the standard deviation is 10 ticks. This represents the data volatility. When you enter a trade, regardless of what the price is, you should expect the price to move at least 10 ticks. What of course you don't know, is whether it will move up 10 ticks or down 10 ticks and whether it will do so in a linear fashion. It may move up 5 ticks and then move down 10 ticks. The point is you shouldn't expect the market to move say 20 ticks after your entry.

Thanks. This is actually exactly the thing I'm asking. Why would the market move 1SD? In statistics, that's just an easy way to reference dispersion, which in a normal distribution, represents around 68.2% of all data around the mean. Besides this artificial construct, it has no significance in itself. In other words, in markets, what's special about 1SD?

 

For example, let's assume this presents an edge in trading markets (I believe you that it does). However, if we used randomly generated data, there would be no edge. Otherwise, you could devise profitable casino betting schemes based on movements in your P/L in games that have a house edge (and ask any statistician if this is possible).

 

So, for there to be a market edge, there has to be reason that markets tend to move in 1SD increments (even if the reason is: they just do, we have no idea why). Here's several charts of data. Some is randomly generated, some is market generated. All edges fail on randomly generated data, but if valid, generate an average profit on market generated data. The question I'm getting at is: Why does the market data provide an edge, where the random data does not?

 

 

Solution key, use Rot13 to decode:

enaqbz: bar, gjb, sbhe

znexrg: guerr, svir

[BlowFish 10]

Do we know random data does not provide an edge? What about random price & volume data? (As an aside it is usually really easy to spot random data when random volume is included).

 

I think the answer may lie in how markets behave. Accumulation / balance / congestion followed by trend / range extension / mark up? Basically the stuff that makes market data not random.

[atto 20]

Do we know random data does not provide an edge? What about random price & volume data? (As an aside it is usually really easy to spot random data when random volume is included).

Yes, we actually do. This is probably outside of the scope of this thread, so respecting jperl's work, it might be smarter to open a new thread for that.

I think the answer may lie in how markets behave. Accumulation / balance / congestion followed by trend / range extension / mark up? Basically the stuff that makes market data not random.

Exactly. That is why Auction Market Theory, support / resistance, etc works. People create markets, and people leave behind biases at times.

[jperl 11]

Thanks. This is actually exactly the thing I'm asking. Why would the market move 1SD?

So, for there to be a market edge, there has to be reason that markets tend to move in 1SD increments (even if the reason is: they just do, we have no idea why).

 

That is my thesis for the threads Atto. If you know the standard deviation, you know what to expect for the market movement. Is this an edge? I don't think so, since you don't know a) when the market will make this move and b)you don't know market direction on the move. You would have to have some additional info to make a decision.

[firstnatt 10]

The skewness is 2nd distribution and kurtosis is third . Value of -+0.847 represents symetric distribution.

I have skewness v1 indicator for metatrader , and market statistics v4, which are great,

but if someone will be so nice to provide kurtosis indicator and perhaps more advanced

skewness indicator.

[akifusenet 10]

Hi firstnat,

 

I wrote the market statistics and skewness indicator for mt4. (I am speaking specifically for forex) I am not using skewness indicator:roll eyes: and for the all statistical computing you can look at this one.

 

Std Deviation - Page 2 - Forex Trading

 

And since jperl is still answering questions I want to thank thim for these great threads. For my experience in forex with market statistics, it can give you definitely an edge but it is still not an easy job. And this is definitely not a holy grail and I belive there is no such thing.But this can really help you.And I am saying this even if there is no real volume in forex and all these calculations use volume and this can make use of market statistics in forex flawed but i believe it still works.

 

And i would definitely advise you add your pivots(or should I say HUPs) onto your chart. They are invaluable for forex. If you dont know anything about forex at least add pivots to your chart.

 

I hope I can keep up:)

 

Take care.

 

Akif,

[firstnatt 10]

Thanks Akif,

 

In 4 hrs charts in Market statistics v4 indicator, how many days back should I input?

35 days ? or 20 days . What number will be precise?

 

It make quite a differance.

 

Thank You.

[enochbenjamin 10]

Jerry this is by far the best thread I have ever come across. You are really a rare breed - answering questions for years without asking for anything in return. You are the real deal and thanks so much for posting your ideas.

 

I came across this thread after having a horrible week (this past week) after CME changed the tick reporting for financial futures. I read the entire thread Wednesday night, watched the market all day Thursday and traded it live on Friday with outstanding results.

 

I already have market bias based upon rising/falling POC's (MP) and extensively use HUP's (virgin POC's, Pivots) as areas to trade, but your methodology provides more information and thus confidence to enter the trade.

 

I will be re-reading over the weekend but just felt it necessary to say thanks!

 

All the best to you.

e.b.

[akifusenet 10]

Thanks Akif,

 

In 4 hrs charts in Market statistics v4 indicator, how many days back should I input?

35 days ? or 20 days . What number will be precise?

 

It make quite a differance.

 

Thank You.

 

4 hrs is smtg i never tried. I m usually at 1-5 min scale:)

 

Akif,

[firstnatt 10]

I guess J. Perl is using 2 min. time frame (market statistics v.4).

 

Perhaps Institutional investors are on 2 min charts.

 

 

[enochbenjamin 10]

I am pretty confident that Jpearl stated several times thru out this thread that time was irrelevant with market statistics.

[akifusenet 10]

I am pretty confident that Jpearl stated several times thru out this thread that time was irrelevant with market statistics.

 

Exactly.Those lines doesnt change depending on timeframe.But if you are using a 4hr chart then you would have only 6 bars for one day and then you should start your calculation for example beginning of the month or smtg but if you are using 2 min time frame then when you start your calculation from the start of the day you have enough data for 2 min timeframe.

[firstnatt 10]

Hi Akif,

 

Will be good idea to invent the predictive indicator in to the future showing the degree of the narrowing and expending the difference between VWAP and PVP, which it will show the selling/buying power of large institutions.

[abboman 10]

That is my thesis for the threads Atto. If you know the standard deviation, you know what to expect for the market movement. Is this an edge? I don't think so, since you don't know a) when the market will make this move and b)you don't know market direction on the move. You would have to have some additional info to make a decision.

 

Hi Jerry,

I have been reading but haven't posted for quite a while.

So I guess the edge comes in trading when the price action is above or below the VWAP(Considering you have a large enough skew), and the greater odds that it will keep going in this direction.

[be the ball 10]

Wow, spent the weekend powering through all 11 of these threads. Good stuff and nice of you to take the time to put it out there.

 

Just going to throw a few thoughts out there spanning the whole realm of your threads. Please note that despite how it might appear, nothing here is a criticism of the work you presented. Quite the contrary. I am merely looking to spur a bit of conversation on this thread...

 

1) I felt you were a bit dismissive of Market Profile theory early in your discussions but to be honest, your entire course here is simply a logical extension of MP based on the ability to extract more information from the data than ever before. Much of Dalton and Don Jones more recent work is quite similar to what you present here. Had realtime VWAPs been available when Pete was creating the outline of MP.....

 

2) You also stated that the normal distribution was quite uncommon in markets. Not sure I agree. The market is constantly forming normal distributions over various timeframes. Just look at the Tues-Fri distribution in the ES for example.... One of your threads discussed trading when the VWAP is near the PVP, which is basically suggesting we have a relatively normal distribution. I would say from experience that we spend more time looking for trades in that scenario than in the scenarios where we have a significant skew.

 

3) Granted this was tailored to beginning traders, but the "Shapiro effect" is a rather primitive way of looking at the price action for confirmation of entry. You state that your timeframe doesn't matter but then base trade entry on a spot below or above an arbitrary timeframe candle. I am surprised no one here has discussed the use of footprint charts showing volume trading bid vs. ask as a "quicker" method of confirming entry on a trade setup.

 

4) I must admit I haven't paid much attention to the dates of the threads but there seems to be a lot of work being done to try to create VWAPs and SDs in various platforms. I have been using MarketDelta charts for about 18 months now and have always been able to get realtime and historical VWAP and SD lines on my charts, calculated off of tick data, non CPU intensive. Just throwing that out there as seems like lots of folks working hard to get what is easily available...

 

5) This goes back to the very beginning. Why are trades in the direction of the skew the safest trades for the Newbie? Rephrasing, why is the high volume area the higher probability side to trade? You led off with those statements but I don't believe you ever stated the reasoning why? Of course, you are correct, but why?

[BlowFish 10]

Wow, spent the weekend powering through all 11 of these threads. Good stuff and nice of you to take the time to put it out there.

 

Just going to throw a few thoughts out there spanning the whole realm of your threads. Please note that despite how it might appear, nothing here is a criticism of the work you presented. Quite the contrary. I am merely looking to spur a bit of conversation on this thread...

 

1) I felt you were a bit dismissive of Market Profile theory early in your discussions but to be honest, your entire course here is simply a logical extension of MP based on the ability to extract more information from the data than ever before. Much of Dalton and Don Jones more recent work is quite similar to what you present here. Had realtime VWAPs been available when Pete was creating the outline of MP.....

 

2) You also stated that the normal distribution was quite uncommon in markets. Not sure I agree. The market is constantly forming normal distributions over various timeframes. Just look at the Tues-Fri distribution in the ES for example.... One of your threads discussed trading when the VWAP is near the PVP, which is basically suggesting we have a relatively normal distribution. I would say from experience that we spend more time looking for trades in that scenario than in the scenarios where we have a significant skew.

 

3) Granted this was tailored to beginning traders, but the "Shapiro effect" is a rather primitive way of looking at the price action for confirmation of entry. You state that your timeframe doesn't matter but then base trade entry on a spot below or above an arbitrary timeframe candle. I am surprised no one here has discussed the use of footprint charts showing volume trading bid vs. ask as a "quicker" method of confirming entry on a trade setup.

 

4) I must admit I haven't paid much attention to the dates of the threads but there seems to be a lot of work being done to try to create VWAPs and SDs in various platforms. I have been using MarketDelta charts for about 18 months now and have always been able to get realtime and historical VWAP and SD lines on my charts, calculated off of tick data, non CPU intensive. Just throwing that out there as seems like lots of folks working hard to get what is easily available...

 

5) This goes back to the very beginning. Why are trades in the direction of the skew the safest trades for the Newbie? Rephrasing, why is the high volume area the higher probability side to trade? You led off with those statements but I don't believe you ever stated the reasoning why? Of course, you are correct, but why?

 

1) The key thing is that the method Jerry presents is based on stats whereas MP is based on heuristics. It's discussed at length in a couple of places.

 

2) Just cause it is bell shaped doesn't mean it is normal I think I am right in saying if the data is skewed it cant't be normal? Thats leaving aside fat tails and such like. (So even if it is not skewed it is unlikely normal with the tails that you see in financial time series)

 

3) Indeed the Shapiro effect is a simple price action trigger based on a previous bar breakout. Other triggers could be used. What is interesting is the circumstance Jerry tended to use it and those he did not. You could use market stats on a faster chart for a trigger if that is more appealing.

 

4) Open a chart of a couple of months of single tick ES bars with a VWAP and SD bands on and see how quick it calculates and loads . It is pretty easy to produce indicators that work fine they are available for a whole load of packages. It does not matter if you 'average' VWAP really anyway. Im sure a lot of charting apps do Coming up with a continuous algorithm is a different proposition altogether. PVP is a different matter, that was solved too. Anyway if you read more carefully you will see what the actual issues are and how to resolve them.

 

5) In a nutshell its a 'with trend trade', so safest for newbies.

[be the ball 10]

1) The key thing is that the method Jerry presents is based on stats whereas MP is based on heuristics. It's discussed at length in a couple of places.

 

--Um, I trade MP and do it from a completely stats based perspective. If you consider MP to be a blind adherence to a 30 year old book, that's a pretty silly belief. That'd be like basing your life's decisions on a strict interpretation of a 2000 year old book. ;-) The way I and countless others trade MP in this day in age is no different than what Jerry espouses here. I think we really have a matter of semantics here...

 

2) Just cause it is bell shaped doesn't mean it is normal I think I am right in saying if the data is skewed it cant't be normal? Thats leaving aside fat tails and such like. (So even if it is not skewed it is unlikely normal with the tails that you see in financial time series)

 

--Again, just semantics in my book. We spend more time fading the edges than we do going with a trend simply because we are rangebound more than we are trending. When we are rangebound, the VWAP, the PVP, the POC and the VPOC are all usually in about the same spot.

 

3) Indeed the Shapiro effect is a simple price action trigger based on a previous bar breakout. Other triggers could be used. What is interesting is the circumstance Jerry tended to use it and those he did not. You could use market stats on a faster chart for a trigger if that is more appealing.

 

--Yes, agreed. I use a footprint chart with delta.

 

4) Open a chart of a couple of months of single tick ES bars with a VWAP and SD bands on and see how quick it calculates and loads . It is pretty easy to produce indicators that work fine they are available for a whole load of packages. It does not matter if you 'average' VWAP really anyway. Im sure a lot of charting apps do Coming up with a continuous algorithm is a different proposition altogether. PVP is a different matter, that was solved too. Anyway if you read more carefully you will see what the actual issues are and how to resolve them.

 

--LOL, pretty sure I would never need to calculate VWAP for 3 months at the tick level but I see your point, and I get that there are a lot of programmer types on these boards. My point was intended to be... that time would be better spent in front of an order flow screen. No matter how cool or accurate your indicators are, you can't be successful at entries and managing your trades without extensive experience reading order flow.

 

5) In a nutshell its a 'with trend trade', so safest for newbies.

 

--Yes, agree.

 

 

See my thoughts in the quote above

 

Thanks for the response.....

[natedredd10 10]

1)

2) Just cause it is bell shaped doesn't mean it is normal I think I am right in saying if the data is skewed it cant't be normal? Thats leaving aside fat tails and such like. (So even if it is not skewed it is unlikely normal with the tails that you see in financial time series)

 

Normal distributions and markets are not just wrong, its knowingly wrong...Which to me is an alternate concept for being "lazy"..While of course if we take a large enough sample size we can find a bell shaped distribution in market data, it has nothing to do with the actual distribution, its a function of how much market data there is...

Just like it wouldn't be hard to find a Poisson distribution or any other distribution you want to look for..you could do this with the data on the size of a single blaid of grass if as much data was being collected on blades of grass as it is on single tick market data.

For whatever reason, the brain just loves to fight that something as complex as the markets are based on something other than an unstable distribution..an unstable distribution with a shitload of random jump process going on.

What Jerry outlined here is a great filter if you are as good a price action trader as Jerry..if you take this as some sort of science then you are screwed because it has nothing to do with reality.

Does Jerry even post anymore? Maybe he got his head taken off during armageddon trying to trade of the variance of the wrong distribution.

 

come out come out wherever you.

[BlowFish 10]

Well put Nate.

 

I am not sure how you "trade MP from a completely stats based basis" betheball? If you would care to describe that then we might have something to discuss. The thing is MP has absolutely no valid statistical basis.It is completely heuristic. Of course that does not mean it "doesn't work" and if it works for you thats great. If you want to kid yourself MP has some statistical basis that's fine too. Anyway I look forward to hearing about your "completely stat based perspective" perhaps the discussion can move on then.

 

"--Again, just semantics in my book."

 

Then you have re written the book to support your view (which you have not actually described yet). It is clearly not 'just semantics'. I guess we are unlikely to have any meaningful dialogue with such different views. Financial data series are not normal as can be proven by employing any one of a number of statistical tests (I would guess Jerry might prefer Shapiro-Wilk ) or simply looking at the distribution.

 

I would recommend you read the threads a bit more carefully most of this stuff has been discussed.

[tradepro 10]

Hi Jerry,

with the longer term HUP's such as 5,10, and 20 day,do you use a rolling time period or a fixed time?