While trend following is tough in intraday timeframes, the strategies have been shown to work consistently in longer-term timeframes. Why don't you overlay a simple trend system with a longer tempo? On a diversified portfolio, buying on the highest high of the last 20 hours/days/weeks and sellling/shorting on the lowest 20 hour/day/week low (or any similar strategy) may not make huge returns, but will probably have a positive expectation over the long run. Furthermore, the CT strategy will most likely make the most profits when the longer-term trend strategy is in drawdown, and vice-versa. The result? A less-steep but smoother equity curve. Your CT strategy is a "short volatility" strategy (lots of small gains and few big losses). A trend strategy is "long volatility" (lots of small losses and a few huge gains.)
Hi drm7, thank you for the inspiring input. Yes you are right. Coordinating the two components CT and T (trend system) is actually what motivated the latest efforts and assessments to determine which T games work best with CT. In fact, i felt that past data provided too little information (and danger of <b>"curve fitting"</b>) to that purpose. My current "panel" for strategy rules has evolved in something as shown below, where i also added the T component. <img src="http://www.datatime.eu/public/gbot/StrategyRules.jpg" /> On the right you can see some simple settings to control the <b>"Trend system"</b>. Now your emphasis is essentially on timeframe and, i might guess, also on independence of the trending system. The reason why i am attempting to make work the Trending component dependent on the CT is because i am looking at it essentially as an <b>"hedging" device</b>. Why am I doing that ? Well if a trader is trading CT only, <b>he can ** in any case ** also trade with another system, independently, which works with trend</b>. The point of putting together CT and T in the same system seemed to make more sense to me if the Trending game could somehow interact with CT to offer protection when actually needed. In fact, as you correctly point out, CT by its nature tends to greedily grab relatively smaller profits, but because of that it may be draw down by adverse price move. At this point T should enter to protect, while not spoiling the CT game. Your other point is about timeframe. At early stage of this development i have been thinking a while what "time" should mean in the context of trading. Should we really care of "absolute time", or what matters is actually the <b>"price movement" occurred during that time</b> ? For instance, a distance equal to 2 "scalp sizes" could be covered in 1 minute or, in a slowly market, in 2 days. Do do we really care about this time interval ? I somehow resolved that my "watch" should have a "clocking" mechanism based on "price alone". Don't' know if you have seen this video here http://www.youtube.com/watch?v=KHjpBjgIMVk&feature=related there is a photon clock. Somehow it's like if we used the "price" in place of that photon ;-)) Therefore, i use to "slice" the price curve (those green lines you see on the trading window) and take <b>"spaced decisions"</b> based on price movement, to be sure to have strict control on PNL. For sure, one could let in any case let a trend system run independently. That would provide a more generic sort of "protection", not necessarily "in sync" with the countertrending entries. Which may still be pretty good. Tom
If it's a true random walk there is no special advantage to buying at new lows or selling at new highs. Let's assume that the log of the price undergoes a one dimensional random walk, (the assumption made by Black Scholes), so prices never go to 0. If you are at a new low the price is just as likely to keep going down as it is to go back up. The past history of the price tells you nothing about where it's going to go next. That's what is meant by "random walk". All successful trading strategies must find *some* way in which the price deviates from true randomness, and take advantage of that.
The paradox you reference makes use of a dependence between the two games that is due to the probabilities of one of the games being based on the capital you have. You probably won't find very many real world prices whose motion is dependent upon the size of your trading capital, particularly in the modular way that it is in Parrondo's paradox.
Right rew, thanks a lot for the interesting insight. It's true that this is build as a <b>memoryless</b> process and that "If you are at a new low the price is just as likely to keep going down as it is to go back up". However, if you look at the set of <b>paths</b> (and we trade entire price paths), those which reach far (from the initial price) at the end of the timeframe are less probable. This might justify why the statistical results seems to favor that behavior (on the other hand the feature keeps appearing in all experiments... you can test yourself to verify: happy to provide you with a copy of the prg). (There should be no doubt about the randomness of this data. It's simply generated using the uniform random generator built in in the .net environment, and and doing a simple Box and Muller transform. Further, all generators seem to agree.) Tom
Thanks auspiv and rew. Very interesting contributions indeed. Well, without arriving to logical paradoxes where 2 losers are turned into a winner, I suspect that in general a "combined" strategy of 2 losing strategy, is not - in general - a "bigger" losing strategy. In fact, even losers can experience <b>"reciprocal hedging"</b> effects. Also, when we get to this kind of subtleties bordering logic, it depends on how exactly "losing" is defined. Tom
Hi friends, I have to give a corrected version of my (random) simulations. In fact, i found an error in the generation process. That was due to the fact that i used the equations generally available on various sources, like for instance: http://www.vosesoftware.com/ModelRiskHelp/index.htm#Time_series/Geometric_Brownian_Motion_models.htm but, after some more careful inspection, i realized that something was not right. In fact, these are not for <b>leveraged</b> instruments, as i am simulating. So the volatility was actually affected by the multiplier of each instruments (thus arriving sometimes at incredible values). Now I have corrected that bug. With the occasion, I have also made the generation <b>more realistic by using a folio of 2 instruments</b> governed by a Geometric Brownian Motion with mean reversion. Each simulation corresponds to <b>more than 70 years</b> of clean tick data for each instrument (1000 replications of <b>truncated</b> sessions of 3 weeks each). <a href="http://www.datatime.eu/public/gbot/Strats G-BOT/Strategy_CT_T_LS/Strategy_CT_T_LS.htm" target="_blank"> Strategy_CT_T_LS </a> <a href="http://www.datatime.eu/public/gbot/Strats G-BOT/Strategy_CT_LS/Strategy_CT_LS.htm" target="_blank"> Strategy_CT_LS </a> <a href="http://www.datatime.eu/public/gbot/Strats G-BOT/Strategy_CT_L/Strategy_CT_L.htm" target="_blank"> Strategy_T_L</a> <a href="http://www.datatime.eu/public/gbot/Strats G-BOT/Strategy_CT_T_LS_E/Strategy_CT_T_LS_E.htm" target="_blank"> Strategy_CT_T_LS_E </a> <a href="http://www.datatime.eu/public/gbot/Strats G-BOT/Strategy_CT_LS_E/Strategy_CT_LS_E.htm" target="_blank"> Strategy_CT_LS_E </a> <a href="http://www.datatime.eu/public/gbot/Strats G-BOT/Strategy_CT_L_E/Strategy_CT_L_E.htm" target="_blank"> Strategy_T_L_E</a> The mechanical strategies above are the countertrending game (CT) with or without "trenders" (T) Long/short (LS) or unidirectional (L) and possibly with a constraint of "extreme entries" only (E). The people who have got my application can clearly replicate all these results locally on their pc. And i take this chance to express my gratitude to all of you supporting this project with your invaluable expertise and constructive feedback. Next, i will resume live testing so that we can compare these various different <b>mechanisms</b> in live testing. Tom
CT T LS results (running from wed) <img src="http://www.elitetrader.com/vb/attachment.php?s=&postid=3065436" /> (image reduced from 1680 to 1000 px width) [If interested in looking at the single trades for each intrument, PM me I will send a complete report.]
CT T LS E results (running from wed) <img src="http://www.elitetrader.com/vb/attachment.php?s=&postid=3065471" />