MtnDog

Thanks for posting this quote. Fascinating observations. I think I need to be more "rat-like" !

Thanks for the replies on targets/trade objectives. I guess there are multiple strategies that could be employed, depending on # of contracts, risk tolerance, etc. As for premium arb, I have been watching this for some time and there are definitely times when the prem hits significant levels and there is no TI. Cash just snaps back in line with futures. Then there are also times when prem arb is happening along with TI. I guess this is more evidence that there are multiple participants and multiple objectives at any given time.

Thank you FT for posting this concise list. It is very close to what I have come up with on my own by reading through these various threads as well as trial and error/ experimentation. My challenge now is to put these inputs together to come up with some kind of tradeable edge that can be programed. This is where I hope some advanced computational methods can help ( ie neural, genetic, data mining, etc.) . I am experimenting with Ward Systems Chaos Hunter right now, with mixed results. Simply by watching these factors in real time, and in a format that my brain can process (akin to UB's HUD), I have developed some intuition and have improved my tape reading skills. Has anyone discussed how they trade using this TI info? Scalps, swing trades, etc.? I think it can lend itself to both, depending on the action you see.

My First, some Wyckoff: "Whenever you study the tape or a chart, consider what you see there as an expression of the forces that lift and depress prices. Study your charts not with an eye to comparing the shapes of the formations, but from the viewpoint of the behavior of the stock; the motives of those who are dominant in it; and the successes and failures of the buyers and sellers as they struggle for mastery on every move." Game Theory: "Expressed in the specific terminology of advanced theorists, poker (TRADING) can be defined as an asynchronous, non-cooperative, constant-sum (zero-sum), dynamic game of mixed strategies. While the game is played in an atmosphere of common knowledge and no player possesses complete knowledge, some players are better able to process this common knowledge into a more complete knowledge than are their opponents" -Pat Dittmar, "Practical Poker Math" For those of you who have never traded in a live pit (before electronic execution became dominant), it used to be relatively simple to figure out trade "intensity" and who were the "dominant" players. If the pit was empty and everyone was reading the newspaper or talking about the cubs/whitesox, then there was no intensity. It was also straightforward to determine what the motives of the big players were as well as order flow, paper (literally), etc. As a local, you paid big money to either lease or purchase a seat in order to gain this advantage. Now, electronic execution has made it much more difficult to determine who is doing what/when. That is why it is necessary to employ advanced methods (neural, genetic, millisecond resolution, etc.) in order to at least get a rough idea of what is really going on in the market (especially in the ES). This is definitely not the only way to trade and make $$. But if I am making discretional trading decisions, I want as much "edge" on my side as I can get. At some point there may be overkill and too much info to process, but that is up to the individual trader's style and temperment ( I knew successful pit traders who barely graduated high school as well as genius/savant types who could count numbers like Rainman). I'll be honest and say that I've learned a lot from UB's posts, but you have to do the work for yourself, and it isn't the holy grail. At least I have become more cognizant of the fact that there is a lot going on down to the tick/millisecond level that I was never aware of. A lot of trading is about confidence in your methodology. Of course there are alternative methods to measure intensity and "commercial" participation. Maybe just a simple candlestick chart with volume is enough for some. I rather enjoy "forensic" trading analysis (ie picking apart trades, volume.bid-ask, up-down tick, etc. etc.) It is tantalizing that all of the information is seemingly right there in front of you, if you can only put it all together in a bigger picture. That is where I think advanced computational data mining/analysis, neural, genetic, etc. becomes necessary. Have a nice Holiday weekend, and take a minute to remember why we have the day off on Monday.

Maybe the second data feed provides something like "Fast Cash" or PREM? This way speculators could be separated from the arbs. eSignal provides EPREM A0, but not to great resolution, maybe down to 1-2 sec. That might get in the ballpark, but not super accurate.

I wonder if there is any use in monitoring the Premium on the same chart as a way of filtering the Intensity readings that are due to arbitrage. A lot of the premium feeds from data providers are kind of slow though. High TICK levels might be another thing to look at. It might be helpful to know the context in which a spike in intensity occurs.

As far as I know, MarketDelta functions down to the second resolution. The Volume Breakdown Indicator has settings for Vol per sec, delta per sec, etc. This gets you close, but not exactly down to ms precision as UB describes. The VB indicator does have something called "trade intensity" which will register a 100 (max) value if the ticks come in one on top of the other. Market Delta recommends this ind. to be used on 1-tick charts for best results.

I am struggling a bit working within the "confines" of a bar. I have been able to code up a dll that can count time between trades with ms granularity, but still need to do more processing within the bar to get significant values. For example, you could take each trade's contract volume divided by the elapsed time since the previous trade, and average those values across the last N values. Or you could accumulate the elapsed time it takes to reach a certain contract threshold. I had some time over the weekend to examine a "spike" down to the tick level. For those of you who haven't done this, it is quite interesting. In particular the spike in intensity on 9/17 @ 11:28.50 EST had the following (approximate) characteristics: Duration: less than 3 seconds # of trades: ~700 # of contracts: ~ 12,000 Distribution of sizes (in bins): 1 lot....311 5 ....211 10 ....56 20 ....47 30 ....20 40 ....7 50 ....5 75 ....17 100 ....3 200 ....5 300 ....1 400 ....4 500 ....2 >500 ....5 Empirically, the "bursts" of intensity seem to happen all in groups of 10-20 trades. The are all "odd" lots (24,3,22,8,17,5,etc.) Sometimes the runs almost add up to a round number (random # generator splitting up the block?). I saw no pattern as to bid or ask trades. The bursts were on side ticks (like a machine gun). Hope this helps.

Thanks. The spike on the low was pretty obvious on my charts as well. I appreciate you opening my eyes to this type of analysis. The concept of trade intensity is sound, and I believe that there may be several different ways of applying/interpreting this information (depending on one's time horizion).

Hey UB, I wonder if your Algos picked up a burst of Trade Intensity at 11:10.43 EST today (9/11). I detected 8 trades all at 1042.50 in rapid succession totaling approx. 1200-1300 contracts (depending on how you split them up). Near the AM highs. I still have some work to do on normalizing for time of day, so I wasn't able to see in real time, only in hindsight. Or maybe was noise/false signal. Thanks.

Thanks UB for sharing your valuable insights. I just recently found this forum, so i have much catching up to do. A question regarding your weighted bias index. Can you give any info about what you are optimizing the weights 'for'? In other words, are you simply optimizing the weights for whatever combo results in the most profit? Or are you possibly using optimized weights that maximize the correlation to a momentum function of price? It also seems to me that this would be best accomplished via some kind of neural net/genetic algo rather than the generic "out-of-the-box" optimizers included in most trading software. Actually, a quick and dirty method would be to simply sum up the weights (-1,+1) of the 24 biases, forming an index that ranges from -24 to +24.