If I understood you right, your system can generate profits from completely random price data, and this would even be an ideal situation. Does this mean that you can take your system to Las Vegas and always win in Roulette? Roulette returns are pure random walk curves, aside from the 'zero' drift that might be the equivalent to a commission.
LOL. No, you did not understand right I merely talked of statistical "strategic dominance". (Which might also mean that, if S is a loser, S* can still be a loser, which has a statistical property to lose less.) You can't hedge, "superpose players" and layers, have free order sizing and spacing, use correlations, and "preserving information" in a Casino setting would probably sound like a sort of frown-upon practice, and you may end up not very well ) https://www.youtube.com/watch?v=xax9SMyeoFA Casinos are designed (and enforce strict policies) not to allow systematic money for anyone else but the house, and they are a form of mere entertainment, that you pay for. But you may certainly try and let us know what are the other differences you find
Yes, I understand that any edge can be only statistical - and the casino proposal, well, was more a thought experiment . Still, a system with a statistical edge even when trading random walk curves is probably unique on this planet, if not in the universe.
Skepticism is always a good attitude for a scientist. And yes, this is (currently) the only approach with player "superposition" and preservation of past trading information. But I expect these concepts and architecture to catch up and become a basic prerequisite in the near future, as professionals will catch up with the ideas. [ I always strive to be original: following herds and being an educated parrot has never been much for me ] In fact, with the naive scheme Open/TakeProfit/MemorylessStop [OTPMS] (which I often refer to as "single-threaded") and under random walk, there is obviously no way to exit from the situation of a continuous PNL fluctuation around 0 and negative drift (due to trading expenses and spread), which makes ordinary trading appear more like the gambling and entertainment in the casino, than a scientific endeavor. And any drift people see in backtesting within that scheme is just a "curve fitted delusion" (expecting flames on this ) ). As a researcher, I can't accept a purely gambling approach, and that's why this perspective was developed.
I must remain sceptical about a positive drift from trading random curves, but I can certainly agree that most backtests results are delusional because of curve fitting. BTW, you're not the absolutely first one who preserves past trade information. I'm doing this also in a system, but only for a special purpose, for recovering losses due to undercapitalization. This allows trading with less capital. My method works anyway only with certain instruments where the price movement is artificially restricted, f.i. EUR/CHF. It would not work with random price curves.
I am glad to hear that. Schopenhauer and Haldane evidently got it right: "All truth passes through three stages: First, it is ridiculed; Second, it is violently opposed; and Third, it is accepted as self-evident." "Theories have four stages of acceptance: i) this is worthless nonsense; ii) this is an interesting, but perverse, point of view; iii) this is true, but quite unimportant; iv) I always said so."
Well, not only you need statistically acceptable performances under suitable stochastic processes, but I will put it much more bluntly: "The fact that a trading methodology must be able to statistically cope with a(ny) "realistic" random process is a prerequisite (with no guarantee of success) to be considered at all." Were "realistic" is defined here as a random generator of price curves such that no "expert witness" could exclude any of its realizations or their distribution from actually happening in the reality. Surprisingly enough, many algorithmic strategies used do not fulfill (not even try) such basic prerequisite. And, even worse, some people obtain curve-fitted parameterizations, by scrambling (searching) the parameter space, but keeping fixed a few past realizations, thus completely reversing the logic of a meaningful statistic simulation analysis. This is, of course, plain nonsense under a statistical perspective. But often this nonsense is sold as "quantitative methods", and herds of people fall in the conceptual trap. The reason why you can throw away any trading methodology which fails to be acceptable under random processes is very simple and understandable. First of all, at macro level, the mkt will be indistinguishable from a realization of a "realistic" stochastic process (this is simple due to the above definition of "realistic"). At micro level, the participants' orders are processed and prices influenced by algorithms especially written by people like me, whose purpose is just to maximize their profit in statistical terms, which means, more brutally, to "rip off" as many participants as possible. So a random process is just and "eden garden" of kindness and neutrality, compared to the "hostile" reactions of the algorithms ruling the actual market. To use a metaphor, if you can't even survive in the eden garden, forget considering living in "sin city". Any trading methodology not fulfilling the above basic prerequisite is a curve-fitted delusion (by definition, as it will only work acceptably on the considered past realizations).
I believe that the opposite statement is true: "It is principially impossible for any trading methodology to statistically cope with a random walk price curve. The existence of such a methodology would cause logical paradoxes, such as generating money out of nothing." Real price curves are doubtless generated by stochastic processes, but they are no random curves, otherwise no trade system would ever work. First, they follow no Gaussian distribution, but a Levy distribution - that's the "fat tails". Second, they have "inefficiencies", such as temporary trends or cycles that allow a limited predictability and can be exploited in a trade system. I'm not making this up, but can prove it - you can easily see those cycles by analyzing real price curves with signal processing tools. But they cannot be found in random curves. Here's for instance the spectral distribution of the S&P 500 index: In a 'real' random curve the spectrum would be flat.
> Real price curves ... follow ... a Levy distribution You are dreaming. Those are just simple probabilistic models proposed by humans, not an intrinsic property of the mkt. > Real price curves are doubtless generated by stochastic processes, but they are no random curves This sentence makes no sense to me >Second, they have "inefficiencies" "Inefficiencies" respect to what ? As I see it, it's far more likely that the observer's mind is "inefficient", than the mkt >signal processing tools There is no such a thing like a "signal" in the mkt. That's made up by scarcely creative humans, borrowing from physics. > "It is principially impossible for any trading methodology to statistically cope with a random walk price curve. The existence of such a methodology would cause logical paradoxes, such as generating money out of nothing." Money is not generated by nothing. Capital, risk, trading information, tickdata, energy, hard work and knowledge, incorporated in the trading process is far from being "nothing". Anyway, we can happily agree to disagree and I stand with my statement: strategies not able to cope with "realistic" (as defined above) random processes are (by definition) curve-fitted delusions. I know it can come as a disappointment to you, but that's just a common sense fact to me.
What are inefficiencies : http://en.wikipedia.org/wiki/Inefficient_markets What is signal processing: http://en.wikipedia.org/wiki/Digital_signal_processing I am not disappointed. I accept that science or logic has often a difficult stand especially against bizarre theories and misconceptions circulating in the trading scene. So, I wish you success with your miracle system - after all, your chance to end in profit is almost 50% .