Quote from itmediaco:
I have found that investigating the book in real-time is phenomenally fruitful to understanding traders behavior before price moves occur. You could look at static properties such as the distribution of limit prices, distribution of volatility, cancel/replace frequency, frequency of price following bid/ask, energy at each price level. And you could combine that analysis with dynamic properties such as price elasticity, response to order flow, etc.
I have yet to find a product that can do that, so I developed my own. The problem with 99% of retail trading software is that its only input is OHLC/volume based on immediate values (no sliding window).
What you need is real-time software that can 1) collect raw transaction data from your feed, 2) aggregate transactions over different length sliding windows filtered by MM, 3) integrate over time, and 4) display the output in a dynamic visualization that enables either your eye to extract patterns, or with spatio-temporal correlation (Bayesian network model, etc).
Good post.
This demonstrates that yourika is humor.
You do have to use a sliding window. I recommend +/- 9 ticks off the BBid/Bask.
You also need to go one level of consideration farther with respect to time, players and the four levels of play.
At any time only one of the four levels is controlling and this is determined by contract net velocity on two time measures on each level. By splitting the measurement data into about 7 groups, it is easy to sweep the panel and know at all times the time of the next BBid/Bask translation.
L2 exhibits patterns just as does price and there ae under 10 such patterns that sequence in specific orders.
The concept of the sliding window is central to be able to extinguish past unwanted data from the history bank you need to maintain. you must have a kill feature in your look back. This means the window stretches and shrinks somewhat on the backside as time progresses.
In an other thread recently commenting on a presumed error, the person was actually on the verge of discovering the true potential of the markets to deliver capital from pools. I usually profer a 3X the daily range as a likely profit yield; you can immediately observe, as did the commentator, that the multiple is much greater.
For contrast to what you are working towards, Google Steenbarger and Greenspoon who are operating in a vaccuum and facing the wrong direction.
It is a difficult concept to deal with, but once discovered, understood and made continually measurable, it behaves just as a bow wave would for a vessel. You may wish to check out bow nautical design and also the methods of bow release surfactant distribution that is used to optimize hydraulic laminar flow. You have to put the stuff where the problem is.
L2 has to be combined with other data feed to achieve what is possible. To you this may be obvious. It takes significant coding to be able to deal with the player numbers and the four level's independent frequencies of what you call cancel and replace.
When given the choice of pattern viewing for the moving window, go graphic rather than a numeric matrix.
Good luck. Again yourika is just humor for beanies.