Page 1 of 2

Brainstorming Alternative Bar Types and Sizes

Posted: Thu Jul 20, 2006 12:19 pm
by aspTrader
https://www.gotomeeting.com/join/263325172
Conference Call: Dial (402) 756-9010, access code 263-325-172

The objective of this moderated chat is to explore and brainstorm about the features, benefits, and drawbacks of using various alternative bar types for discretionary and system trading.

Emphasis will be on comparing and contrasting the strengths and weaknesses of each BarType with the further goal of determining which BarTypes might be used together for benefit.

At the beginning of the discussion, a method will be proposed as a basis for comparing alternative Bar Types.

Sunday July 23, 2006
5:00am Pacific Daylight Time (currently GMT minus seven)

Connection information will be posted at the link below just before the meeting as usual.

http://forums.neoticker.com/showthread.php?t=1057

Post any questions by sending me a Private Message at this forum.

Cheers!

Posted: Thu Jul 20, 2006 10:58 pm
by michal.kreslik
Great! I'm looking forward to it much.

I've moved this topic to the "aspTrader's NeoTicker webinars" forum and left the shadow topic in the original forum.

Posted: Wed Jul 26, 2006 4:45 pm
by eudamonia
aspTrader,

I didn't have the opportunity to attend this webinar, however, I've read several of your posts on the NeoTicker forum regarding the your interest in range bars. Along these same lines you had several valid criticisms regarding tick and volume bars.

Recently I read this article on "Market Time Data" which is a time data series that responds to market volatility. This increases the accuracy of technical analysis and seems to more clearly define the underlying data without excess noise and without data corruption issues. Unfortunately it is properietary but at least it seems to support the theory that better methods can be developed.

http://citeseer.ist.psu.edu/levitt98market.html

I'd be curious to see what your thoughts are on this.

Edward

Posted: Wed Jul 26, 2006 5:12 pm
by michal.kreslik
Ed, aspTrader,

let's continue in the "brainstorming alternative bar types", do a U-turn and think about this:

My opinion is that the bar construction techniques are in fact indicators.

What is the definition of an indicator?

I'd say it's a mathematical method to use some market data set to transform it into another data set.

This way, the various bar types are nothing more than indicators that take the tick data set and transform it into another set of data (the set of sets of 4 values - OHLC - per some predetermined set of tick data, like per 5 minutes in 5 min bars or per 20 price points in price-range bars).

Now this is really an eye-opener: this all means we are actually applying the indicators to another indicator (artificially constructed bar) when we place a new indicator on anything other than a tick-chart.

Every indicator distorts the price action in some way, although it's true it sifts out some noise, too.

The real question here is: why are we so obsessed with using these indicators (bars) over and over again? We may develop strategy code that sifts the noise out in some better way than by using this indicator (bar).

Using bars this stubbornly is in fact the analogy of using any other indicator (for instance, the moving average) in every single strategy code you write regardless of the strategy's inner logic.

Michal

Posted: Wed Jul 26, 2006 5:49 pm
by eudamonia
Michal,

Interesting point about tick data.

In my opinion, however, price without context to time is a highly confusing and noisy data set. I would argue that time as well as price is important in making financial decisions. After all, even comparison to tick bars utilizes time on the x-axis to make decisions.

What are some ways that we could evaluate price independently from time in a meaningful manner?

Edward

Posted: Wed Jul 26, 2006 7:21 pm
by michal.kreslik
Ed,

I am not talking about evaluating price independently from time. Tick data are more time-precise than bars.

Posted: Wed Jul 26, 2006 7:51 pm
by eudamonia
Michal,

That is true, but raw tick data is unfiltered. At some point you will want to apply filtering of the data in some fashion to provide meaningful analysis. This can be represented as a bar or any other mathematical function as you mentioned earlier. All filters will tend to skew your tick data set in some fashion (they ignore outliers, become skewed from missing/incorrect data etc.) How would you propose to get around these issues?

Posted: Wed Jul 26, 2006 8:56 pm
by michal.kreslik
Ed,

what do you mean exactly by the raw tick data is unfiltered? Do you mean raw = just received from the data feed provider?

Of course, we can only work with correct data without the glitches and meaningless values. If there's a nonsensical value like a 300 pips spike between two ticks, this very spike would certainly show in the bar constructed from such tick data, unless we perform some kind of filtering.

If you mean filtering as a way to diminish the noise, then there are plenty of ways on how to do that. For starters, why not apply the common-and-garden simple moving average to the source tick data? You would then come up with the data set that is not jittered and still have more information than the bar indicator - every "averaged" tick would feature its original time value.

Taken this above example, you can assess whether the new high (of the averaged ticks) is a new high of the last true 15 minutes. Quite surprisingly, you can't do this with any time-based bars. Why? Since the ticks are boxed into predetermined start time - end time intervals. You simply don't know when exactly and in what order the data came in. Especially with Tradestation (thanks God I'm out of that :) ) where the indicators cannot Look-Inside-Bars.

If we have second-precision stamped tick data, we can asses very precisely whether the current averaged tick is the highest one of the LAST 15 minutes (meaning, for instance, 14:03 to 14:18 ).

Most importantly, this way, you have indefinite number of "bar closes" where you can enter/exit a trade since the monitored condition can occur at every averaged tick. (well, it's not indefinite, it's limited by the number of averaged ticks you've got available, but compared to the standard bar indicator adjusted data, it's truly "indefinite")

Michal

Posted: Wed Jul 26, 2006 9:08 pm
by eudamonia
Michael,

Sorry for the confusion. I wasn't understanding how one would apply an MA to the tick data. I've been stuck "behind bars" too long :) Thanks for clearing that up.

How would you put something like this together in NeoTicker? Do you need to create a special indicator to keep track of time?

You've really opened my eyes on the possibilities of time-stamped tick data. Thanks.

Edward

Posted: Wed Jul 26, 2006 9:23 pm
by michal.kreslik
Ed :D :D

"stuck behind bars" - that really is an apt remark :)

Obviously, if you apply SMA to the bar indicator, you are losing even more information since in the most cases, there's only one price used for calculating the SMA indicator (most often the close price).

I don't know how would I put this together in NeoTicker yet. But I guess I'll know that by tomorrow this time :) Since my NeoTicker package has left Paris already (I am tracking it with the FedEx tracking service - am I not crazy:) ), it will be here in the Czech Republic tomorrow very probably :D :D :D