Regarding negative autocorrelation of the returns of trading systems, I believe it. My current trading system (which is based on reversion, not trending) also displays tendencies to have winners after losers and losers after winners. But how you capitalize on this observation, especially under conditions of shifting market behavior, is another issue. Playing games with position sizing can increase your performance, but can lead to high risk when the price pattern changes and you find yourself placing a very large position on a very losing trade. BTW, not all fractals are random. The notion of a fractal is simply a characterization of a pattern or phenomenon across a range of scales. It says NOTHING about the mechanism that generated that pattern. Some fractal generating mechanisms are highly structured and arise in highly deterministic conditions (e.g., the branching growth patterns in biological systems). Trade carefully, Traden4Alpha P.S. The quote is from "The Predictors" by Bass, an excellent book for anyone thinking about trading system development, especially automated trading systems.
I do not pretend to be an expert in statistics or probability. I could never get my arms around the gambler's paradox. I understand the tosses are independent but what about the probability of successive heads or tails? Anyway, to the problem at hand, namely can you increase the performance of a system by waiting for "n" successive losers before taking signals? It seems to me that there is a basic fallacy at work. Even if we were guaranteed that we would get more winners, it is irrelevant. We shouldn't care about percentage of winners/losers. We care about net profit. Simply saying , or even proving, that you were less likely to get a loser after a run of losers says nothing about the size of the winner or loser. You would have to try to model the size of winning trades after various distributions to get any kind of useful results. I have actually tried to do that, but gave up. It was obvious to me that the risk of missing big winners far outweighs extra losers. Another approach is to trade the equity curve. You start trading a system after a statistically significant drawdown. That has intuitive appeal to me but again, you run the risk of missing a big positive run.
AAA, You make very good points, which is why I stated previously that "it is an advanced topic". Often times the small gain in EV is not worth the time/effort/psychological impact in adding that extra layer of position/money-management. You really need to look at overall EV you are getting from applying that method in relation to changes in return-on-risk. Also, you can adjust position size using these techniques rather than skipping a trade entirely, because that trade signal should still be a +EV trade. Perhaps some real-life examples would bring some clarity to the topic, including my own personal use of this technique. When I was a long-term trend follower, I did not have enough capital to diversify across enough markets to truly utilize the Turtle approach. I was able to have open positions in 5 markets on average depending on the market and my equity. To help make up for lack of diversification, I did research on the likelihood of trends to continue after a winning trade of X magnitude (in relation to MFE in past Y trades) specific to my entry and exit signals. What I found was that the market was on average more likely to consolidate (resulting in more whipsaw trades) after a large winner. So what I did to makeup for my modest diversification was rotated markets after certain winning trades. If I was in Soybeans and just had a winning trade that fit my criteria, I would rotate that % of equity to a different market that recently had at least two consecutive losing trades. I think it is quite reasonable to believe that markets have an increase tendency to consolidate after a large trend; and vica-versa. From this observation I did thorough research specific to my trend-following method (similar to Turtles in principle but different entry/exit criteria) and found statistically valid data to implement my diversification model. There are a variety of systems such as band-based counter-trend systems that will stop taking signals after 2 consecutive losses in same direction in a certain time period (say 2 losing trades using 5m bars in same hour period, both playing for bounce off lower band). This is to protect against the likelihood that it is a trending day rather than a range bound one. These systems will have certain conditions that need to be met (such as price touching the other band or some momentum-specific indicator value/xover) before it will take another position against dominate trend. There are a variety of methods to trade breakouts from res/sup that will increase position size after failed trades at the same res/sup area. For example, you trade a triple top breakout but it fails. Perhaps increase position size on the 4th or 5th attempt at the breakout? Did I mention P2's methods, stat arb, seasonal hedging strategies too many times already? Anyways, I am not making any judgements on the general profitability of above methods as the devil is always in the details, but I thought I'd bring some clarity to the discussion by providing specific examples where traders would consider modifying position size, skipping trades altogether, or using porfolio diversification in direct correlation to the result of a previous trade(s).
Hi, My initial post is just 26 hour old and I would like to thank all thread participants for their contribution to this discussion and the various links to books, websites & ET threads.. (didn t have time yet to explore all of them) As I initially said, I have no idea if using a system like : only take trades âafter having experienced two losersâ would create a new system with a higher profitability than the initial one. I am fully aware that such a system would be missing major move, and that it would be very difficult to execute it according to the book⦠Anyhow, my main concern behind this example was to know if there were some proved theoretical concepts showing that âsuch technicsâ would help to design systems with a higher profitability, ie I am targeting a higher % of winner and at least a similar ratio AvgW/AvgL (of course, I will get less trade, but I am looking anyhow for efficient filtersâ¦) By âsuch technicsâ, I mean more widely : Taking into account the sequence winner/loser (& eventually the size of the corresponding PLs) into a system design⦠It seems that one of the issue is to know if trades are independent from each other or notâ¦. Ie is the market âRandomâ or not ? (not sure this is the correct wording however..) As a 16 month old daytrader (focusing on the ES since last April), here is my personal feeling on that : Letâs consider the ES on a 120 or 180 ticks charts, and a trading system based on entry points which are picked up randomly ( the exit signal would be based on a trailing stop for example.., ie exit long if ES ticks lower than Higher High since Entry less 2 points). In this case, I could imagine that the sequence of winner/loser might be ârandomâ as well, ie one could not benefit from the information "there have been three losers in a row" to improve its original system design (by changing bet size for example) On the other hand, letâs consider a classical breakout system (ie go long when the higher high since 20 bars is broken by one tick, resp go short etcâ¦..) with the same kind trailing stop exits⦠Then I could imagine that this kind of trading system may include some kind of typical winner/loser sequences because these sequences are induced by some ârecognizableâ and repetitive market patterns. For example, an uptrend channel with some bull flags ending with clean breaks, may result in a series of : Winner (big one), series of small Losers (showing congestion, ie the market action do not produce swing bigger than 2 ES pt), Winner (big one) etc⦠Of course, this wouldn'tât show up so nicely, but I just want to say that one could take into account such sequences (if detected) in order to identify some trades to be filtered out or have their bet size modified⦠So I would say I agree pretty much with Dottom initial message on the thread⦠Anyhow, I will try to do some backtesting runs to see if such technics may help improve the efficiency of some of my pullback and trend âdetector filtersâ (that I may apply on such breakout system..)⦠I think that finding a global (and efficient) bet sizing rules (or trade filtering rules) to be applied all the time will be quite hard, so I will mainly focus on the impact of such rules when activated in specific trading situations like for example : increase the bet size (and tighten stop) after x losers if some congestion period (inducing these losers AND confirmed by other "notrend" indicators) is detected⦠Eddy PS : Dottom, I am just reading your above post after having finished mine : you just hit the target when you mentioned these specific examples : this is exactly the kind of situations I am looking forward to backtest and where one could expect some added value of strategies including some "reactive" position size modifications..
My recommendation would be to first make sure that whatever method you are trading is profitable on it's own using standard money-management (i.e. say 2% of equity risk per trade). If you are backtesting just use fixed size per trade. If you still have a profitable system afterwards then you can explore the benefits of position-size/trade management above-and-beyond your original method itself. Some methods you will find no advantage by adding variable position-size/trade management. If you are looking at a trading method that is a loser using same size per trade, but is a winner if you add variable position sizing on top of it, I would be very careful about trading that method. While in theory it is possible to change a losing method into a winning method with position sizing (this assumes you believe that the markets are not random and results of subsequent market movement is influenced by prior movement), it is a very dangerous game as you have very little room for error. You must also be very certain that your statistical analysis is valid and not a result of luck or other market bias. Just my thoughts... good luck!
Just my .02, increasing position size after a drawdown is completely nuts, period- akin to playing a long term game of russian roulette. As Eckhardt notes, flip a coin enough times and the outlier streaks just keep getting longer and longer ad infinitum. The same thing applies to winning and losing streaks for traders- i.e. the longer you trade, the more freaky streaks you will endure, both for you and against you, simply as a matter of course. You can look at your "luck" as a probability distribution- the longer you stick around, the more time you have to explore the tails. So: if you have been leveraging up into drawdowns over hundreds of trades for years and years, long term probability is actually working against you rather than for you, and you are setting yourself up to catch a bullet in the back of the head. The correct thing, in my opinion, is oddly as unpopular as the wrong technique is popular: to increase your position size into winning streaks. This may hurt you a little when those streaks are cut short- but the cost is light, just a shave off the top after already being ahead. And when you do catch an outlier streak in your favor, and have been gearing up to take full advantage of it- that's when you blow the doors off. And you can rest easy knowing it was an extra chunk of profit at risk, not initial capital. Never let them get too deep into your pocket. Ever.
Good points. Likewise, I've cautioned many times that varying position size in this way can be very dangerous if one is not mindful of return-on-risk, but there are some very practical ways to keep such techniques in check. And just as you can increase position size on trades after failure, you can also decrease position size or skip a trade altogether (such as my example of skipping subsequent counter-trend trades to avoid trading counter-trend during potential trend day). As always, thorough backtesting will give you the necessary insight. If increasing position size, the correct approach would be to set practical limits akin to a stop-loss in an individual trade. Such is the case with many stat arb strategies. Perhaps P2 would care to comment on his risk management when a position starts dropping against him with no bounce in sight?
for some reason I can hear the guys at LTCM saying this in soothing tones to their clients, circa early 1998....
Are you seriously using this P2 kid as an example for position sizing/risk management? Matter of time, just a matter of time... Thats like using Kerry Kollins as a poster boy for sobriety... You booooys like Meeehhhhiiiiiiiiiiiiicooooooooo? Whhhhooooo!