Yeah, jack's good at killing braincells. Meanwhile, in the real world, here's my take on genetic optimizations and how I use them. Multicharts has a perfectly functioning genetic optimizer, and here's I use it. We've found that inside of 5 variables that are the key core to our system, we have to find a range of those from 1 to 1000 or 1*10^15=1000000000000000. We break that down into boxes of 100 steps. That is, 1-1001 has a step at 1,101,201...901,1001. Genetically optimizing those ranges we are able to narrow down our ranges further in smaller and smaller, more precise steps. Our optimizations tonight gave us a 1-151, 51-151,1-101,1-101, 1-101 as a starting point. So we started stepping by 50's, then 10's, then 5's then 1's, whereas in the first one genetically optimizing 1,51,151 and all of the other possible values. The end result is that we went from 1*10^15 possibilities to 120,960 possibilities optimizing genetically. Once you deduce the range of values given by succesive short lived genetic optimization, you can then perform your exhaustive studies. Our range values optimizing genetically goes from 1000000000000000 to 120960 which is thousands of orders of magnitude easier for a computer to analyze. What I've realized learning from this process, is that if you understand genetic algorithms as computerized evolution, you can find the perfect set of parameters in quadrillions of possibilities. Where my system stops its evolution has more to do with its automated capacity to self-optimize. I believe the OP is looking more for genetic optimizations built into the strategy automatically than the manual process I used tonight to find 120k possibilities from 1*10^15. Multicharts is beautiful for doing this, and the software packages necessary to build a true genetic robot are simply too cumbersome and next to impossible to ever get to a point that you would deploy them on a server. Getting to that stage is a much bigger problem than a solution, since Multicharts off-the-shelf genetic optimizer can do this for you. Assuming you have a program capable of genetic evolution, the process of narrowing down possibilities to a computable amount that won't take 5,000 years is the whole reason genetic optimization was created, and I'm saying that there is no reason to build your own packages when the off-the-shelf technology, particularly Multicharts, can do this without the additional headache of compiling data, and coding an actual software package to both analyze and implement. Both analysis and implementation is way too much to ask the typical newbie genetic programmer how to do. The pitfalls of using this is that many do not have strategies suitable for these methods. If I were to ask you what the probability of there being another planet like earth with similarly sentient species, the probabilities are about as large before genetic algorithms were created as the time it would take to compute 1000^5 possibilities on a PC. Next to supercomputers, that many possiblities is impossible to compute, but if you're smart you'll use genetic optimization in boxes if you have more than 3 variables to optimize and are able to narrow your ranges to a reasonably short period of time.
And all this time I thought I must too stupid to trade, because I had no friggin idea what Jack's been saying ! Maybe I'm just stupid enough!
Braincell mentioned that, to him, most of my significant words have ten meanings. You can get out of his dilemma by using the glossary I have created. Fortunately we just heard from bow wow. He has a 120K problem. You can avoid that too by using your mind to deal with a finite market and its characterisitics. So far, I have usedsome words NOT in my glossary. For those, use the dictionary's first choice of meaning; ignore alternatives. All markets are NOT continuous in nature. All market, instead are only as finely divided as the price and vo9lume unit defined by the owners and operators of that market. Continuous functions cannot be used in markets as a result. Someone suggested running a created system until it breaks. In this concept, the finite nature of all trading systems appears. All systems have parts. They stem from participant tools or from the market. All my parts can be spread out for viewing quite simply. For the market I use three things: 1. price units, 2, volume units, and 3. Participant's activity: accumulation or distribution. I have several tools. They were designed by their authors. The tool and the anuthors are: 1. Paradigm Theory, Keynes, 2. Logic Theory, Carnap, and 3. Boolean Algebra, Boole. To look more closely at the tool sets, I will list the important tools in each set. 1. Hypotheses, and parametric measures, 2. And, Or, and compares for equality, less than and greater than, and 3. True and False emerging from the circuits of 2. just above. The youngest age group that I have found that can handle all of this is fifth graders. The consequence of the application of these things to the market are three "one pagers". You can see several independent systems superimposed on the one pager that illustrates the parent from which all applications are derived. It is "The Pattern". On the drawing you see "scoring" that comes from the three market variables. You also see the parallelogram showing the trends being stacked up on interlocking fractals. 8 parts, which is 2^3 was to be expected. To be clear, the fact of granularity was the prime motivational source and granularity dictated the rest of the deduction. Arriving at the ONLY and simplest and most elegant solution. was certainly NOT rocket science. so, to be sure, you can read what I write either as clear concise and crisp or you can choose "gibberish, drivel , goobly-gook or any other term you wish. That is your priviledge or as a red neck would say, your right. It is my desire that anyone who works can learn to build his mind to have a fully differentiated spectrum of differentiation. What this means to the unlearned is that the mind, through drills, can be grown to read the market just like a person drives a car. Not doing the drills precludes this possibility. Anyone can ask: "Have I done the drills?" if the answer is no, then the person cannot understand what I write. to drive a car, you do drills to learn to drive. when you have done the drills, you can ask:"Can I drive a car?" The answer will be Yes. I do not know if bow wow can drive a car but I do know hye has not done the drills ever in his life. He is spending his life looking for a shortcut.
The problem is that people decide when to buy or sell. So the question is; what external events cause a majority of people to be buying or a majority of people to be selling? Can *all* of those external events be found and can all of those external events be input into a computer? Perhaps the words "market crash" in a national news story would cause a lot of people to sell. On the other hand, the words "market crash" might be interpreted by some big money folks as "Blue Light Special for next 30 minutes in entire store", so they go buying up the bargains. Anyway it seems to me that news stories play a big part in the direction the market takes on a given day. So the idea would be to monitor all those news stories. But... It takes several seconds to just load the NY Times main web page into my computer when I click on their site. Imagine a computer trying to read just the headlines from many different news outlets at the same time! Let alone also download and read stories which might be of interest. Quite a bit of bandwidth would be needed to say the least! Anyway IF you could constantly download all those news stories and market events just as soon as the stories were posted, then you could set your evolutionary software loose on that data and it could learn what makes things tick. But... If you are constantly loading main pages from a particular news site, I would imagine they would notice a large amount of bandwidth going to a particular IP address and then block it. Possibly see it as a "denial of service" attack? So I would think you would need to pay for news wire feed services? And then there is the question why when Ireland has financial problems does Intel stock go down because of that? Or why does the entire U.S. market go down because Greece is having trouble? Does someone sit at home and read the morning paper. See that Greece is having money problems. Then call his broker and tell him to sell all his Intel stock?
I found the NASA stories... "Thirty five networked computers running advanced evolutionary algorithms, generating hundreds of thousands of antenna designs." http://www.nasa.gov/centers/ames/news/releases/2004/antenna/video.html "NASA Ames scientists used the software to 're-invent' the antenna design in less than a month to meet new specifications - a very quick turn-around in the space hardware redesign process." http://www.nasa.gov/centers/ames/news/releases/2004/04_55AR.html
Oh yeah, I thought everyone knew that. But let's not forget that market action can be analogous with the hydrological processes and interactions of an enclosed fluid body within parameters of shoreline wherein inputs of wind direction and strength interact to cause waves of action that then reverberate off shoreline and bottom. These waves do not disappear but continually interact on an exponentially decaying basis These waves can then, from within the chaos of interactions, superimpose themselves one upon another or cancel each other out. Thus from seeming chaos can there be synchronate activities resulting in peaks and troughs that the trade-wind sailor can use to hasten the journey. Undercurrents can often be deduced by their actions upon the visible waves. The peaks and troughs belie an inherent consistency within the chaos and by considering the trend of the inputs one can harness these forces to help progress one toward the anticipated course.
not to thread-jack...but i got a question.. I'm fairly familar with RapidMiner and VERY familar with R....I'm always curious how people trade with these tools..Do you have a custom c++ code that calls these programs (if possible) to generate a signal? Thats the only part i cant quite solve... I'm pretty sure there is no interface for these programs from brokers, only API (which i've come to find out is more difficult to learn than i thought). I've got good R knowledge and good rapidminer knowledge, but little C knowledge (damn, lol)..
LTCM had an edge but was overleveraged and was liquidated before making a huge profit. If you do not overtrade you run no risks like that. or get a system with a very high edge and watch it RTM. See this thread for detail: http://www.elitetrader.com/vb/showthread.php?s=&threadid=215283&perpage=6&pagenumber=3