If possible could you provide a graphic illustration to express the performance of incorporating velocity and acceleration calculations. Also, is there a reason as to why this would be used for commodity trading systems, as opposed to Forex or Equities?

a quick search using the phrase "why are futures better to trade than stocks" yields a host of reasons via your favorite search engine. but there are others as well more psychological.

@Mark, my algos are mostly oriented toward counter trend trades with small profit targets and hold times of milliseconds to a minute or so. I think of price velocity as being like throwing a baseball in the air. At some point, the ball is at the top of its arc and begins to fall; I want to fade the move as close to that point as possible within the constraint of getting the required size.

It is a lot of code but conceptually very simple to find where price has accelerated to an unsustainable point and then has begun to decelerate (although price may still be rising). If you can estimate where that point may be and stack orders there, you will have decent queue priority without being a full stacked book trade. One method of estimating where price will likely begin decelerating is to track local min/max extensions as these tend to be relatively stable especially at new highs/lows for the day.

Interesting - I used tensors back in the 70's for modeling complex vibration problems with deformable materials. Last several years modeling multi-timeframe FX transactions. Would love to collaborate.

Mark,

you state : "backed with statistical foundations it is indeed possible to develop anti-trend systems incorporating physical relationships with accuracies of better than 80% as to identifying the exact price sometimes to the tick) at which a move will terminate and reverse"

Have you actually developed such a system or are you just providing us with an academic, theoretical essay ? If you did develop such a system (80 % accuracy), what has been the live performance and why would you like to share this holy grail ?

However, if you did not develop such a system, how do you know it has an 80 % accuracy as theoretically, the impact of orders you would set to trade and that would impact the price, can not be estimated.

Thanks for helping me understand

I am not a theorist, I am a researcher who actually trades for 3 decades and support top ranked CTA's with trading systems. The 80% model has a independently audited track record for 3 separate funds. It was developed in the early 80's on the Ned Davis Research platform Technalyzer for a single market the 30 year bonds futures. The model utilizes a unique set of switches that flip on and off as a combination of events happens. I will be happy to send anyone a document which outlines the concept. Obviously the system is not the Holy Grail because it endures some large single position drawdowns occasionally affectionately called silver bullets. Trading always in without stops has it's pain but decade after decade of profits sooth that anxiety.

Hi Mark, yes, I would like to see such a document.

Hello Mark, you state : "backed with statistical foundations"

What do you mean by statistical foundations? Which paradigm do you refer?

Calogero Bennici Here is a good roadmap with many branches to climb out on. My road was machine learning and human interpretation, implementation. https://en.wikipedia.org/wiki/Foundations_of_statistics

I would be interested as well to read the document. Thank you

thanks Mark. Looking forward to receiving said document

@ Mark Brown I would be interested in viewing that document.

I would also like to read the document.

Thank you in advance.

I would also like to see the document; every little bit helps.

Thanks Mark. I also would like to see your paper.

7.) don't assume that all data is to be split in half it's ok to detect bias and adjust a model's tendency to perform better in phases from buy to sell side. 8.) don't assume that so called "gambler" indicators or just indicators in general can't produce successful systems. 9.) i learned that systems that look good on a chart don't always make the most money 10.) i learned to reduce contract size on statistically better than average profits spikes . mark

Hi Mark - Thanks for a very intersting thread. I am also very intersted in reading your document.

Scott Boulette you will see that this bond model is a very unique big counter trend reversion to the mean system. i traded hundreds of contracts at a time with it and usually received +positive+ slippage on the fills. something that has to be mastered if your trading institution size slow boats. the model takes a beating drawdown wise when it gets caught up in events but usually it reverts. when it doesn't it takes the loss and moves on, in fact i learned to like losses so we could move on back into the dependability cycle.

i don't not dismiss trends though they exist and the reason you will see a reversal routine in the bond model code. at sometime when well underwater it will finally release the switches and flip "if" the momentum is persistent and at lest attempt to recover some of the previous position loss. hope this can benefit you in your counter trend trading. m

Hi Mark. Interesting reading. Would appreciate seeing your document. Thanks.

Hi Mark, Would be interested to see how you have formulated this. Thanks

Mark, I appreciate the information. In my particular case, I have lots of metrics (not indicators per se but more involved than simple calculations) that allow me to mostly stay out of event driven moves. I spent years working on algos that detect adverse selection events. On their own, they don't do much but when fed into other algos, they make a big difference. If you want more detail send me an email (you have my email address).

Hi Mark, Also Interested to see that document. Thanks

Mark, I would appreciate the opportunity to read your research. Thank you

Mark look like a pretty interesting approach. I would appreciate if you can forward me the doc describing your research.

Thanks,

Prasad

David Rosa for some reason I can not send to you? All others sent, thanks!

Hello Mark, your approach sounds very interesting indeed. I would love to see your research on this.

Talking about reversals , these Dollar And GDXJ charts might help SEE future ; http://borisc.blogspot.com

Hello Mark, i would appreciate seeing the document. Thank you.

Hi Mark. Interesting idea, would appreciate reading your research. Thanks

Hi Mark, sounds like we've been barking up the same tree so the speak. Would love to compare notes.

Hi Mark, I'd love to read your paper. Thanks in advance.

Hi Mark, interesting area of investigation. Would love to read more if you are still distributing your paper. Many thanks.

Mark, I am interested if your paper contains charts.

Hi Mark, I would love to read your research on this. Thanks.

Hi mark, it would be great to read your article

I wanted to explain a little logic about model building in general. There is always a fight between Trending and Reversion trading methods. They both have their strengths and weaknesses I will not go into right now. But lets accept that both methods have their faults and that usually when one is winning the other is losing. So each method will set on the sidelines or even endure some losses waiting to get into sync with its style of trading.

Consider designing a model which has a trend method and a reversion to the mean method combined into one trading system. Most markets chop more than they trend, so we would focus on building the best reversion to the mean system we could and ignore the big losses that occur when a trending trade comes along.

Now armed with a well tested and satisfactory reversion to the mean systems we can implement rules to stop and or stop and reverse a reversion trade when it is outside of its statistical average. m

To sum up we would assume every top is to be sold and every bottom to be bought. We would run with that assumption until the point that both momentum in the opposite direction along with a stop loss amount that is greater than our 75 percentile winning trade size is. At that moment if we have hit our stop amount and the momentum is rising still against the position, the model will flip and go with the trend direction.

If the momentum to reverse is absent the model can either stop out or stay in the trade. Switches flipped on and off in the code allow a multitude of variations to run simultaneously becoming a very dynamic systems that can handle diffrent types of market conditions. m

@Mark - there is another model category that is largely unknown/ignored/dismissed and which rarely has any significant losses, certainly not over a period even as short as a week.

If you have very complete, clean data, reasonable processing power, decent latency (ultra low latency is not required) and a good commission structure, simple order flow/volume analysis for very short term trading with small targets is a way to avoid the issues you mention above. I am not talking about stacked book hft trades (although they qualify) that are beyond the capabilities of virtually all individual traders.

The primary issue becomes technology and a solid set of statistics on various scenarios. This approach requires advanced order handling and a method of identifying market conditions that are adverse to this type of trading (when there is doubt, there is no doubt). Even given those requirements, this is still achievable by the individual trader.

@Scott - I learned allot about latency while officing at the CME building on the 8th floor. We had a fiber cable run from the members server in the basement up the elevator shaft to the office at considerable expense. I was able to compare latency of the CME member feed against Globex, TT CQG, Esignal lol, and feed provided by Mizuho bank.

The delay in real time should be prosecuted as a crime. Honestly and with all respect I think you maybe taking for granted where you are as applying to traders across the country. There was a reason I flew to Chicago every couple of weeks to trade. I did not dare try and trade that style trading from Texas even setting on a OC192 in the Telecom corridor of Richardson one block from Alcatel. You are very lucky to be where you are the delay spreads from there.

Trading styles are something that Scott brought up and I think it's a subject that is basically ignored so I want to talk a bit about it. I have done everything from HFT to waiting on Options to expire on a monthly basis. Granted I don't have much real world experience outside a 30 day trade this is a fact. You MUST find your frequency it's is as important as having an edge to trade. Much of the frequency size is dependent upon an honest evaluation of where you're at geographically and how much time you have to commit. I believe that many out of exchange area traders in an effort to reduce losses will try and trade down on a tick level. Truth is if you drive a truck you should be not trying to race the Indy 500.

Large institutions have size requirements that make developing systems with positive slippage a must that individual traders may not need at all. I know of many traders and firms that are successful trading only End of Day - "I wish I were one".

@Mark--What you've described is precisely what I do, except the machine learning aspect. I refer to the two sides as outperformance and reversion. I look at changes in instantaneous or continuous velocity also. Interesting.

Is there any chance you could direct me to some resources regarding machine learning. Or, better yet, I'm in Chicago also, and perhaps we could meet for coffee?

Also, Mark, have you noticed an asymmetry between both the magnitude of a move and the accuracy of signals (buy versus sell)? I trade many products with this method and certain exhibit this asymmetry.

@Timothy thank you for your comments and reach out. I travel to Chicago less than in the past but still get up there. Absolutely but not brute forced I have found asymmetry in most everything tradable including option selling. It's slanted world we live in lol, but I am only an observer of the fact and capitalize on it. I spend most my time studying how to apply that discovery being very careful to not over compensate and step over the threshold of curve fitting. There has to be a good fit but some flexibility "at the cost of profitability" else you will destroy long term dependability. It's a bit of art but heavily rooted in science principles and procedures during analysis.

@Mark, I trade from my home office (not in Chicago) and have a vpn into the servers in Chicago. The dashboard (to get an overall sense of where things are) and a mechanism to trade manually or manage a position in a more sophisticated manner if necessary is all I need locally.

The execution management is all in Chicago while the analytics server can be anywhere. By fully distributing the various components, I can manage any latency issues. However, ping times from here to Chicago are probably about the same as you have, so as long as the signals are not dependent on sub millisecond latency and the execution management is microseconds from the exchange, the location of rest of the components doesn't really matter. Something like redis pub/sub is good enough for communications among the components even when they are not on the same machine; it works great for anything other than true hft.

I guess in your terms, I drive a really fast sedan.

@Scott I hear you brother but I hate that VPN stuff haven't resorted to that in twenty years lol. I can tell this now cause I'm not doing it any longer. We were trading FX interbank arbitrage with servers at NY1/2 and LD1/2 across the pond we were trying to figure out how to reduce latency to London. We tried everything and then it hit me how to reduce it close to zero. I would send the entry signal first and then the exit signal in the same order maybe milliseconds apart at most, lol it worked like a champ. Latency conquered, however some people were not happy at all.

@Mark, the vpn is required for a secure connection. The traffic going up to the execution server is very light, just command and control stuff. What comes down is just data needed for position/risk management. And yes, if crossing the market, I always send the exit first.

Lol, apparently we learned a lot of the same stuff, in the same way, at about the same time.

Scott Boulette yes it's funny the lessons that run parallel when you're putting forth maximum effort, many come to the same conclusions. it's that effort and often failure "in my case anyway" that tempers your opinions about methods of trading i guess.