Pardo Robert The Evaluation And Optimization Of Trading Strategies, Second Edition

Podstrony

[ Pobierz całość w formacie PDF ]
.It certainly goes without saying that this would bean excellent trading strategy.If, in fact, you have one like this, you do notneed this book.Go now and make your fortune.The alternative view is that to achieve optimal trading performanceover long periods of time with a trading strategy, it will need to be up-dated, at least occasionally.It is also part of this view that it is per-fectly acceptable for a trading strategy to employ different parameters forc06 JWPR070-Pardo December 18, 2007 14:17 Char Count=The Historical Simulation 141different markets.This is, in fact, preferable, in that it adds an additionallevel of portfolio diversification.Making this case in the necessary detail is an argument beyond thescope of this book.Suffice it to say that I have seen a lot of trading strate-gies over the years and I have yet to see a trading strategy of the foreverand one-size-fits-all markets, trends, and conditions.In fact, it has even been said by Richard Dennis, of Market Wizard andTurtle Traders fame, that the principal driver of the Turtle Trading systemno longer works.3 But worthy of mention on this note is that the TurtleTrading system and variants thereof are reputed to have produced hun-dreds of millions of dollars in trading profits in the years when it did work.Also, I have seen profitable real-time trading strategies that do use dif-ferent parameter values for different markets and do benefit from periodicreoptimization.In fact, the models employed by the commodity trading ad-visor Pardo Capital Limited fit this profile.Its performance is a matter ofpublic record for those who are interested.I suspect that if you were able to look behind the scenes at a numberof professional money managers, you might be surprised at the degree towhich their models and strategies have changed and developed over time.As a matter of note, the current climate among large institutional in-vestors is that a money manager should be constantly looking to change,adapt, and develop his strategies.Quite a change from the one-size-fits-allcrowd of a decade ago.It is the operating assumption of this book that models and even strate-gies will benefit from modification from time to time.It is also one of thereasons that Walk-Forward Analysis was originally created.WINDOW SIZE AND MODEL LIFEWe assume that the parameter sets of the majority of trading strategies willrequire and benefit from the adjustments produced by reoptimization.Wenow must examine the relationship between the size of the test windowand the duration of its use.And by usage, we mean two things.The firstapplication is an out-of-sample window in a Walk-Forward Analysis.Thesecond application is the length of time a model can be traded in real timebefore requiring reoptimization.Many trading strategies, particularly those that are nonadaptive, whichbenefit from optimization, will require some type of periodic reoptimiza-tion.The effect of this reoptimization is to adjust the trading model tocurrent market action by locating the best model parameters for currentmarket conditions.c06 JWPR070-Pardo December 18, 2007 14:17 Char Count=142 THE EVALUATION AND OPTIMIZATION OF TRADING STRATEGIESExperience has shown, and common sense agrees, that a trading strat-egy optimized on a larger test window is likely to last longer between reop-timizations.Conversely, strategies optimized on shorter test windows aremore likely to require reoptimization more often.Models built on smallerwindows are said to have a shorter life.Those built on bigger windows aresaid to have a longer life.The main reasons for this are structural.The time between reoptimiza-tions has generally been found to be some smaller portion of the originaltest window size.This is highlighted by Walk-Forward Analysis, which ispresented in Chapter 11: Walk-Forward Analysis.A good rule of thumb forthe determination of the size of the trading window is to set it at betweenone-eighth and one-third of the test window size.For example, if a 24-month test window is used to optimize the tradingstrategy, then it can safely be traded out-of-sample or in real time for be-tween three (24/8 = 3) and six (24/3 = 8) months with minimal likelihood ofadverse performance.Let us try to provide some intuitive concept for thisprinciple.Consider the following example.A strategist optimizes a strategyon two years of data.Would he be confident trading this model for the nextsix months? Probably.Would he be confident trading this same model fornext three years? Maybe not and with good reason.This is somewhat analogous to the application of a similar statisticalmodeling procedure.It is known that statistical forecasts are the most ac-curate for the next period forward and less accurate for forecasts madefurther out in time.Let us assume, for example, that we are using a poly-nomial regression model to forecast the closing price of oil.The forecastmade for the next period out will be more accurate (and with narrowerconfidence bands) than a forecast made five periods out.This will be dis-cussed in more detail in the next section.The main reason a trading system must be reoptimized is that marketschange with some frequency.They do not change, however, with great reg-ularity.If market conditions remain the same, a new reoptimization of thetrading system will most likely arrive at the same parameter values thatwere identified by the previous optimization.Conversely, if market condi-tions change, especially dramatically, reoptimization will most likely iden-tify new parameter values.So, what is the main reason why a short test window is most likely tobe tradable for some minor fraction of its size? Primarily, because a smalltest window is unlikely to include a comprehensive sampling of markettypes and volatility levels.Consider that a small window where the major-ity of market activity is a strong bull market can be said to have seen onlyone type of market.Review Figure 6.12 for an illustration of this concept.Also, if it has been a fairly consistent bull market, it is relatively likely thattrading volume and volatility have been relatively consistent as well.c06 JWPR070-Pardo December 18, 2007 14:17 Char Count=The Historical Simulation 143FIGURE 6.12 Chart of a strong one year bull marketWhen a trading strategy has been optimized on a test window with asteady bull market and stable volatility, it can be said that this model hasbeen adapted to this type of market and set of conditions.In fact, one couldsay that the model has been particularly well adapted to such conditions.Perhaps the strategy is too well adapted.The strategy will perform wellunder similar conditions but it is unlikely to continue to perform well if themarket trend changes from a steady bull market to a choppy bear market.When market type and volatility change into something that the strat-egy has not seen in its life, there is no guarantee that it will continue to per-form as it did during its previous optimization.Reoptimization empowersthe trading strategy with the capability to adapt to just this type of chang-ing market and unfolding conditions [ Pobierz całość w formacie PDF ]

zanotowane.pl

doc.pisz.pl

pdf.pisz.pl

bezglutenowo.pev.pl