Quote from TSGannGalt:
http://www.elitetrader.com/vb/showthread.php?s=&threadid=164155
http://www.elitetrader.com/vb/showthread.php?s=&threadid=161203
Oh gawd....
OK... jack... start a thread in the Automated Trading Forum and I'll go head2head with you. Just PM me with the thread link... I don't think this is the area for me ask you questions and doubts about what you provided regarding your trading method....
Trust me. I won't be nice and this IS a "AUTOMATED TRADING" discussion. Why don't you start off with posting your test results with "Cash Cow" that the "so called trolls" are emphasizing... jack... no charts... send me those codes for "CashCow" and I'll test them for you.
Seriously... again... I don't like your words... I'm only going to deal with codes and Real CashCow that you keep on emphasizing in those ATS threads. Anyways... you can start trying by replying to the links above...
I guess this is your one critical question:
"jack,
You really haven't answered anything... You know that right?
OK, let me re-ask the questions, I'm guessing that my questions weren't done correctly. Maybe I've been scattering questions too much:
Q1. Please provide an explanation of the theories for your 3 models and the supporting proof of it.
Q2. Please provide a log (preferable *.csv) of your trade results. Preferably, if you can also include the statistics used to analyze the trade log would be helpful. My tester can analyze the viability of a model based on it's trade log so it's not big hastle. It would help though, if the format is (It's not a bid deal even if you don't):
yyyy/MM/dd hh:mm:ss.fff, symbol, price, contractsize
... single trade per line, my app will match the orders, even if it's legged trades, on it's own. If it's a US exchange traded symbol, I usually have data for it. It can handle any timeframe, also. If you have your models under ATS, you must keep a log of your trades, right. I just need a snippet of them... reasonable size to validate the significance statistically though."
My response:
We use a different type of log. The biggest difference is we include the trade direction information and what the signal source was, meaning the logic source(s). All the trading addressed here is in one futures market as you may know. We also have many more columns than you use. Use the one pager for PVT and SSR.
The basic core was end of bar trading so the times are just multiples of 5 minutes.
The seasonal shift was noticable on the version being done and after the time of your critcal question. There are only about 150 trades a month and per contract the monthly variation was 4400 to 1400 as summer set in. We shifted the income by refinements. A monthly shift of 2800 was noticed on the 1400.
For question 1. refer to the P,V Boolean relationship and its parametric measure.
Let me repeat it here to save a lot of work:
H1 If volume is increasing, then price trend will be continuing.
H2 If volume is decreasing, then price trend will be changing.
The parametric measure is a binary vector. The four measures are:
1. increasing and 2. decreasing for volume. They tell you what is going on now for price trend.
3. continuing and 4. changing for price trend. This is a result and it is measured to determine the answer to "What is the right side of the market?"
The basic core measures the channel of price.
The first shell measures the channel as well using degrees of freedom that allow for the connection of a series of bars to address the "order of events" in a cycle of price and volume. As you know a cycle of a variable, as measured in binary vectors, has many aspects. I'm sure you recognize the normal displacement has its derivatives than can be measured one upon the other using a binary approach as applied to non-continuous functions. displacement, velocity and acceleration each have direction and magnitude. Two other shells are devoted to traverses of channels and tapes of traverses.
In addition to determining what event in the order of events; it is good to know how fast or slow events are occurring in terms of bar times. The above provide such an early warning system.
We also know "What Must Come Next".
I would make this simple for you too by adding another hypothesis for your benefit. It is not a rigorous requirement since this is already included by deduction in H1, H2 and their combo.
H3 The operating point of the market migrates in its matrix and does not jump around. While some would approach this problem with probability, we do not; we use Carnap instead. (See: logic theory)
This is a case limiter according to the dimensions of the matrix. Our maximum case limit is 16 in the basic core. The deductive test is conflict. You do not test in the deductive manner nor do you include a conflict function nor its resolution logic.
Unfortunately, from what you shared with us, you have an invalid statistical method. (See: inductive proofs).
I know that the matrix test can only yield correponding powers of bases as the series of values, but it is easy to recognize that the binary aspect does more than reduce the series value.
Dealing with "What Must Come Next" is an intellectual exercise that is unsurpassed for learning to trade. It is what shifts trading from checkers to chess. As you see easily on a playing board, only half of it is used for checkers and all of it has possible uses in chess.
The market operating point only moves if ther is only one move possible. So in trading, conclusing waiting is the primary trading function of analysis; the name of this waiting is called "HOLD". anyone who is assessing the quality of trade results is primarily interested in the hold between behavioral actions. Your evaluation cannot do that in any sense.
How does logic examine what is not possible and only leave what is correct in the piture? This is deduction as exemplified by null hypothesis testing.
Hold, in effect, is the answer, over time until all alternative possibilities have been eliminated.
Keynes requires that the hypothesis set be of like kind and the parametric measure be of like kind. This, once established, allows the application of logic theory in either a probabalistic orientation (the orientation you have without passing Keynes tests) or non probabalistic. Choosing to not have to face Bayes or "frequentists" is not a difficult choice. Most people go quant and choose both Bayes and"frequentists". We chose non probabalistic, the opposite of your choice.
You have seen our volatility vs volume nonstationary look up tables. Horizontal and vertical Gaussian distributions abound. (See charts' recent post) Pragmatically we unanchored volume from an axis so it more resembled price which is far from its horizontal axis. Hi and H2 still apply and apply more neatly. geometry can now be used more easily and in "like kind" ways to the variables of the market.
The hypothesis set, its parametric measures and What Must Come Next, completely answer all questions needed to be sked to trade with certainty at all times.
The proof is:
Ask all questions required for "sufficiency", then when all answers are binary vectors, certainty is achieved.
The set of questions define holding and taking timely action between holds. A sequence is used to apply the requirements of sufficiency. Monitor, Analysis, Decide, Act. The most common ction is hold.
Here are some things to notice. You deal inductively and get the consequence. I do not and I am not part of the inductive world; it is simply too risky.
What I did was build something for markets that is like the periodic table is to science. The coe is composed of sub atomic particles called components that are algebraically defined using raw data. The components are combined to form balanced atoms and the atoms make simple and complex molecules and compounds. As in chemistry and biology and physics, there is no need for uncertainty in their Basic use.
Going beyond basics, we design effective and efficient applications as shells that create greater yields all the way to optimum.
What is done is move to faster fractals, fractal by fractal. the effect is that time stretches out and many things can be observed. Human observation is limited to the senses. The PC goe way beyond sensory capabilities of humans.
Pool extraction is the result.
Your Q1 has been answered.
Q2 was an poor inductive question. I chose to not simply back away from it. Instead I suggested to you that the deductive way of doing business is a requirement. If anyone is doing any logging or analysis of logs, the work has to be done deductively.
I haven't gotten to te level of participation where I would start a thread.