NeoTicker Forums  

Go Back   NeoTicker Forums > Trading Related > General Trading Discussions

Reply
 
Thread Tools Rating: Thread Rating: 22 votes, 4.73 average. Display Modes
  #1  
Old 01-24-2006, 11:42 AM
asptrader's Avatar
asptrader asptrader is offline
Regular User
 
Join Date: Dec 2004
Location: Chicago, IL
Posts: 547
The Case Against Tick Bars and Volume Bars

I first posted the note below at the Jurik Tools Yahoo Group back in August 2005. Itís post number 199.

I wrote it before the CME changed itís algorithm for tick aggregation late last year. I believe the note has more relevance today as a result of that change at the CME, especially for system traders.


************************************************** **********
The Case Against Tick Bars and Volume Bars

There are two reasons why I think working (learning) to trade Tick and/or Volume Bars is the worst investment of time and energy OVER THE LONG TERM.

The first thing to say is that I have carefully worded the first sentence of this post. I'm NOT saying that Tick and Volume Bars don't yield good insight and have benefit. I'm merely saying that, given the limitations of time and energy to become proficient and successful with these various bar types, I think working (learning) to trade Tick and Volume bars entails serious problems. I don't have good solutions for these problems and, frankly, I haven't come across anyone yet who has described solutions that, IMO, are satisfactory.

MY MOTIVATION in posting about these issues is to provoke thought and discussion about potential solutions for these problems.

So, what are these 2 problems? In this post, I'm going to sketch the 1st, less significant problem. I'm describing these problems across a few posts because I can't make the time to do this in a single post.

Tick and Volume Bar Problem #1: Every tick Counts!


For both Tick and Volume bars, the trigger for new bar creation is the crossing of some tick or volume accumulation threshold. In a 5000 contract volume ES bar, a new bar is created when the number of contracts exceeds 5000 and not until then. For a Russell 2000 futures 233 tick bar, a new bar is created when the number of ticks in the bar exceeds 233 and not until then.

For both these bar types, EVERY tick or contract traded COUNTS toward the new bar creation threshold. If the trade that takes the number of ticks from 5000 to 5001 is removed, then 1 tick delays the creation of a new bar. Notice also that if tick number 2563 is removed instead of tick 5001 the result is the same: there will be a delay in the creation of a new bar.

Because Volume Bars are formed by transactions that contain a variable and sometimes large number of contracts, the result of removing a single transaction can even more significantly impact the TIMING and (potentially) the PRICE at creation of the next bar.

To summarize: For Tick and Volume Bar Types, every Tick Counts and that's a problem. OVER THE LONG TERM, relying on tick and volume bars ASSUMES a CONSISTENT data feed, correct data cache of ticks, and manipulation of those ticks across data vendors, trading platform vendors, etc. If even a single tick is missing, then the new bar open time (and potentially price) will be different and any indicators applied to that data series could be different as well.

Admittedly, some of these differences are trivial. But they are not always trivial.

Contrast this dependence of Tick and Volume Bars on every tick with the dependence of Time-based bars and Range Bars on specific ticks. The removal of a tick does not impact the creation trigger of a Time-based bar unless that tick is the very first tick of the next new time period. The removal of a tick does not impact the creation trigger of a Range Bar unless that tick is the very tick that constitutes the crossing of the Price Range threshold. In short, dependence on reliable tick data when using Time-based bars and Range Bars is very dramatically reduced and hence, I believe, risk is reduced.

To restate: The expectation of consistent Tick and Volume Bar Type performance assumes reliable, robust, and consistent data feeds and applications manipulating every tick. The expectation of consistent performance for Time-based bars and Range Bars does not assume anything near the same degree of reliability and robustness of the feed or of the application for every tick.


So, that's it. The fact that EVERY TICK COUNTS toward the creation trigger of tick and volume bars is, for me, over the long term, a problem I don't have a solution for and donít want to deal with.

This is especially true when other Bar Types are available that NeoTicker supports which provide significant noise absorbtion.

I'm certainly open to hearing about long term solutions to this problem and hope someone here has one! 8-)

In my next post, I'll describe the 2nd, more significant reason why I think working (learning) to trade Tick and Volume Bars is a bad LONG TERM investment of time and energy when there are other, better Bar Types.
__________________
aspTrader
Reply With Quote
  #2  
Old 01-25-2006, 10:49 AM
William2002 William2002 is offline
Regular User
 
Join Date: Jan 2006
Posts: 61
My interpretation of bar data is just the overall "flow" of price information. You claim that time bars are better b/c a bar is pretty much always going to be constructed at a regular interval. Well what do you do if there are no ticks in that time interval and it flat-lines your indicators. Also, if a break-out price (tick) is missing from a time bar b/c of bad data then the time bar is essentially just as useless as a tick bar b/c it fails to represent "price" movement.
Reply With Quote
  #3  
Old 01-25-2006, 05:45 PM
Support Andy Support Andy is offline
Support
 
Join Date: Sep 2005
Location: Toronto
Posts: 306
An issue with tick bar raised by aspTrader is very true - the tick bars we get here on the testing computers across the data feeds do not really look the same from one data feed to another.

e.g. a 100-tick bar chart for one trader does not necessarily means the same setup or signal generated by another user whose 100-tick bar chart is based on another data feed.

A very odd issue indeed.
Reply With Quote
  #4  
Old 01-25-2006, 11:22 PM
William2002 William2002 is offline
Regular User
 
Join Date: Jan 2006
Posts: 61
Looking forward to the post about other "types" of bars or superposition bars. The description and usage of these bars doesn't seem as straight forward.
Reply With Quote
  #5  
Old 01-26-2006, 12:03 AM
William2002 William2002 is offline
Regular User
 
Join Date: Jan 2006
Posts: 61
I was thinking about what Andy said and if there are conflicting charts for 100 tick bars due to data feed discrepencies, then the time charts are going to be different also. So I would think........
Reply With Quote
  #6  
Old 01-26-2006, 01:34 PM
asptrader's Avatar
asptrader asptrader is offline
Regular User
 
Join Date: Dec 2004
Location: Chicago, IL
Posts: 547
Which is best to sacrifice: Data Integrity or Indicator Usefulness?

William2002:

In your first post above, you state the fundamental problem with time-based bars that caused me not to want to use them a while back. Namely, time-based bars dramatically reduce the usefulness of even the best indicators during small volume periods.

More than that, as I've written elsewhere (and provided chart displays to illustrate):
  • time-based BarType results in relatively more bars when a market is in congestion and relatively fewer bars when a market is moving
  • time based bars provide more close-of-bar strategy entry opportunities during congestion (all the more to get whipped around)

But I disagree that time-based bar OHLC structure is impacted to anywhere near the same degree that tick and volume based bars are. (Because, as is argued above, for tick and volume bars, Every Tick Counts.)

The essential OHLC structure of a time-based bar is only impacted by a missing tick when the missing tick occurs
  • as the last tick of the time period or the first tick of the time period AND that last tick or that first tick is the only Open/Close tick at that price OR
  • as the only tick at the high or low of the bar
Time-based bars: More data integrity but reduced indicator usefulness through low volume periods and "noisy market action"


Support-Andy:

Your post above about different vendor data feeds resulting in different OHLC structures because of missing ticks confirms my own experience with even a single data feed.

On another major trading platform, I have written systems that would have a certain trade list based on data I downloaded one day and then, if I did a data refresh, have a different trade list the following day. For the longest time, I couldn't figure out what the world was happening. That experience is documented here (subscription required for viewing).

I finally figured out what my problem was. It had nothing to do with the platform per se but was entirely rooted in the fact that, for Tick and Volume based bars, Every Tick Counts. The solution I found resulted eventually in me becoming one of the "n/a" posters in that thread.

Tick and Volume based bars: More indicator usefulness through low volume periods and "noisy market action" but dramatic OHLC data integrity structure problems occur regularly and cannot be prevented


Which is best to sacrifice: Data Integrity or Indicator Usefulness?



Until next time...
__________________
aspTrader

Last edited by asptrader; 01-26-2006 at 02:06 PM.
Reply With Quote
  #7  
Old 01-26-2006, 02:13 PM
William2002 William2002 is offline
Regular User
 
Join Date: Jan 2006
Posts: 61
asptrader,

I noticed that you mentioned the CME is now aggregating its ticks. Do you see a solution for this from using a different type of bar? Possibly the superposition bars available from NT? Have you researched this much? Also, under which circumstances would you use "momentum" bars and the like?

william
Reply With Quote
  #8  
Old 01-26-2006, 02:39 PM
asptrader's Avatar
asptrader asptrader is offline
Regular User
 
Join Date: Dec 2004
Location: Chicago, IL
Posts: 547
Quote:
Originally Posted by William2002
I noticed that you mentioned the CME is now aggregating its ticks. <Snip> Have you researched this much?
The most detailed discussion of the new CME tick aggregation method, including how it differs from the previous method, appears to be here.
__________________
aspTrader
Reply With Quote
  #9  
Old 01-26-2006, 03:29 PM
William2002 William2002 is offline
Regular User
 
Join Date: Jan 2006
Posts: 61
asptrader,

Your posted link definitely describes the aggregation in good detail. It is an aggregation, though not quite as severe as I was expecting. Would "momentum" bars be a good alternative to tick bars in this situation? I really am interested in learning more about "momentum" bars and other bar types and would appreciate any insight you have in this area. I know there are several others who may be able to give a good advice on this area as well....so I am open to any suggestions. I don't think there is a "staff" blog on the topic. Maybe I am mistaken?

thanks
william
Reply With Quote
  #10  
Old 01-26-2006, 03:50 PM
Support Andy Support Andy is offline
Support
 
Join Date: Sep 2005
Location: Toronto
Posts: 306
Quote:
Originally Posted by William2002
I was thinking about what Andy said and if there are conflicting charts for 100 tick bars due to data feed discrepencies, then the time charts are going to be different also. So I would think........
The time based charts tend to agree with each other with minor discrepancies only.

e.g. when 2 ticks in one feed is treated as one tick in another, the time based bars will agree with each other, but the tick based bars will look drastically different.
Reply With Quote
  #11  
Old 01-27-2006, 01:02 PM
William2002 William2002 is offline
Regular User
 
Join Date: Jan 2006
Posts: 61
I have been looking at some of the different bar types available with NT, specifically pricerange, vol range, % range, and momentum bars. I am curious what time frame you should work with when using superposition bars. I would think the time frame does not matter if the bars are constructed on "price" or volume data as opposed to time intervals. In addition the bars are spaced with bar interval. Can someone clarify exactly what kind of data should be used to construct a chart of superpostion bars, 1min, 10 tick, 5 min?

appreciate any help
-william
Reply With Quote
  #12  
Old 07-23-2007, 02:48 PM
nelo nelo is offline
Regular User
 
Join Date: Nov 2006
Posts: 19
Asptrader,

what you write is quite true for tick based chart, but not so much true for volume base chart. The difference is that if you use for example 1000V chart and current bar volume is 975 and new 100 lot trade comes, the bar will be 1075 and new bar will start. This adds useful robustness into volume based bars and your system/method will not be influenced almost at all.

Next, one trade is absolutely whatsoever not important in my trading and even if the following bars are then influenced and are different to the correct sequence it again does not in the long-term influence almost any system. The sequence is different but the principles are the same and after few bars few missing ticks are absorbed. The same happens to time based charts.

I have to disagree with your opinion also for third reason. When trading stocks or stock indexes or futures on them you never have all ticks. There are so called dark pools and over the counter trades which will not appear on your T&S, but they influence market (price and liquidity).

Time based charts are in very short term (1-2 short term bars) more robust, but there is a much bigger problem that they change arbitrarily - when new second comes. And change of one second has in my opinion absolutely nothing to do with market activity (mostly, if you do not trade specific news). This makes them very noisy - which in my opinion is why they are the worst investment of time and energy in the long term. That is just another point of view.

Of course, if you have very unreliable data feed tick based charts will be more influenced than time based, but you should not use such data feed for any trading.
Reply With Quote
  #13  
Old 07-23-2007, 03:37 PM
Bruce DeVault's Avatar
Bruce DeVault Bruce DeVault is offline
Regular User
 
Join Date: Jul 2006
Location: Washington DC
Posts: 9,246
The utility of tick or volume bars is much like the utility of any other type of bar compression: whether or not you can make money using them. Different people find different bar compression types useful to them not least because their strategies vary quite significantly from each other (including not only basic approaches but timeframes), thus lending themselves to different methods of analysis; ergo, different people find different bar compression types to be the most useful to themselves personally, and different people's approaches are affected the most by different considerations.

Among other things, major changes in the way ticks have been aggregated at exchanges and popular retail data feeds, together with forever fluctuating (although from a distance, seemingly forever increasing) levels of market activity mean that you can't reasonably ascribe the same meaning to either "1000 ticks" or "1000 shares/contracts" today and 5 years ago (and thus, AspTrader's main points are good ones in the long view), just as in the short term (comparing for instance now with the same time of day yesterday), the comparison can be a relevant one.

Does the fact that you can't reasonably compare tick or volume bars using the same parameters for all time mean that tick and volume bars are completely useless? No (and I don't think that is at all what AspTrader was saying) - it means you need to be aware of what you are looking at, and what types of structural and market considerations affect your comparisons of current data with data from the past. The buck always stops with you - the person in control of your trading. You have to make yourself aware of the issues that affect you and your trading, and AspTrader has done a good job as well as a valuable service (especially for those starting out) by pointing out some of those issues unique to tick and volume bars and thus contributing to the public understanding.
__________________
Bruce DeVault
bdevault@quantevo.com
www.quantevo.com
Reply With Quote
  #14  
Old 07-26-2007, 08:35 PM
jcash jcash is offline
Regular User
 
Join Date: Jan 2007
Location: Monroe and Wacker
Posts: 115
If your missing trade is OHLC on tick or time it makes a difference. Yes.

Another consideration to be made re tick charts...if the tick missing from your bar would have been a duplicate of a previous trade's price, whether it be OHLC or anything in between, it would not have a substantial effect on the overall story that bar was telling. Of course, again, if the tick that took the place of the missing tick within the bar was OHLC then there is a problem.

Missing volume, however, cannot be replicated like a price and thus will always affect a chart.
Reply With Quote
  #15  
Old 07-27-2007, 07:30 AM
Bruce DeVault's Avatar
Bruce DeVault Bruce DeVault is offline
Regular User
 
Join Date: Jul 2006
Location: Washington DC
Posts: 9,246
Quote:
Originally Posted by jcash
Another consideration to be made re tick charts...if the tick missing from your bar would have been a duplicate of a previous trade's price, whether it be OHLC or anything in between, it would not have a substantial effect on the overall story that bar was telling. Of course, again, if the tick that took the place of the missing tick within the bar was OHLC then there is a problem.
I think one just has to be careful what special meaning one ascribes to the open & close of bars. In most cases (smaller than bars of one entire exchange session, which have some special considerations), the only thing that speaking completely broadly is mathematically special about them (as distinct from all of the other trade prices) is that, if nothing is missed (and the compression method does not involve interpolation), they're evenly spaced samples according to whatever rules were used to compress the trades into bars, and that for most bar compression methods (and again, when the compression method does not involve interpolation) the bar's close is the last known trade price at that time - information that can be quite useful indeed, but information that is not unlimited in its scope.

So when one speaks of a missing trade causing a bar to be "wrong", I think one needs to think carefully about what "right" is - especially in a world in which not every last trade is probably going to be seen in an immediate timely manner and in sequence 100% of the time no matter what one does. Does "right" mean your data feed delivered everything it normally receives 100% accurately, on time and in order? Does "right" mean you got even the trades your data feed missed, were reported incorrectly upstream, or are from sources it does not usually report? The situation is even less clear for non-exchange-based instruments such as spot foreign exchange. Clearly, by putting more money and effort into technology, one can get closer and closer to an idealized "right", but it's safe to say nothing is ever always 100% perfect in real-time reporting. Part of trading then is the science and the art of making the most of what is known to be quite possibly incomplete information.

One can and should strive to get the most accurate (and the least noisy) information possible. AspTrader's points are well taken that certain forms of data compression are naturally more resiliant to certain types of missing data (in other words, if you randomly discard a few trades, they're more likely to come up with the same or very similar compressed answers). But no matter what form of bar compression one employs, if one's analysis collapses because of a few missing trades, it should serve as a warning that perhaps the analysis is deserving of greater scrutiny, considering the imperfect world in which such strategies must be expected to function resiliently.
__________________
Bruce DeVault
bdevault@quantevo.com
www.quantevo.com
Reply With Quote
Reply

Thread Tools
Display Modes Rate This Thread
Rate This Thread:

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT -4. The time now is 07:34 PM.


Powered by vBulletin® Version 3.8.6
Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
Copyright© 2000-2010 TickQuest, Inc.