SHORT-TERM TRADING WITH Prick PATTERNS A systematic methodology for the development, testing, and use of short-term trading systems MicHAEL Harris 25th Manivenay 1975-2000Copyright© 2000 by Michael Harris. ALL RIGHES RE- SERVED. Printed in the United States of America, No part of this publication may be reproduced, stored in a retrieval sys- tem, or transmitted, in any form or by any means—electronic, mechanical, photocopying, recording, or otherwise—without the prior written permission of the publisher. ISBN: 0-934380-60-0 Published January 2000 Editing, layout and cover design by Kevin D. Stokes Sth anivenay 1975-2000 TRADERS PRESS, INC.® PO Box 6206 Greenville, SC 29606 Books and Gifis {for Investors and Traders 8222 / 864-298-0222 Fax 864-298-0221 Tradersprs@aol.com Attp://www.traderspress.com Ihe NEA requires the following statement: livpothetical or simulated performance results have cer- ‘ain inherent limitations. Unlike an actual performance record, imulated results do not represent actual trading. Also, since the trades have not been executed, the results may be under- or over-compensated for the impact, if any, of certain market actors, such as lack of liquidity. Simulated trading programs 1 general are subject to the fact that they are designed with the benefit of hindsight. No representation is being made that any account is likely to achieve profits or losses similar to those shown.” The trading systems shown in this book, as well as the pat- terns listed in section three, are for educational purposes only. Past results are not necessarily indicative of future results, and therefore it should not be assumed that use of the rules or techniques presented will result in trading profits. This is not a solicitation of any order to buy or sell. Publisher's note: the System Writer Plus software used to generate many of the tables appearing in this book is no longer available for purchase from Omega Research, Inc. It has since been replaced by Omega Research TradeStation 2000i, a comparable software package with enhanced per- formance capabilities. For additional information, please call 1-800-556-2022.Visit our Website at http://www.traderspress.com + View our latest releases + Browse our updated catalog + Access our gift shop for investors + Read Ed Dobson’s book reviews Contact us for our 100-page catalog 25th Anniversay 1975-2000 TRADERS PRESS, I! PO Box 6206 Tradersprs@aol.com 800-927-8222 Fax 864-298-0222 To my parents, George and Dorothy -M.H.PRADERS PRESS, INC, * Publishers of: A Complete Guide to Trading Profits (Patis) A Professional Look at S&P Day Trading (Trivette) ‘Ask Mr. EasyLanguage (Tennis) Beginners Guide to Computer Assisted Trading (Alexander) Channels and Cycles: A Tribute to..M. Hurst (Millard) Chart Reading for Professional Traders (Jenkins) ‘Commodity Spreads: Analysis, election and Trading Techniques (Smith) ‘Comparison of Twelve Technical Trading Systems (Lukac, Brorsen, & Irwin) Ghlic Analysis (LM. Hurst) Day Trading with Short Term Price Patterns (Crabel) Exceptional Trading: The Mind Game (Roosevelt) Fibonacci Ratios with Pattern Recognition (Pesavento) Geometry of Markets (Gilmore) Geometry of Stock Market Profits (Jenkins) Harmonic Vibrations (Pesavento) Howto Trade in Stocks (Livermore) Hurst Cycles Course (J.M. Hurst) Jesse Livermore: Speculator King (Sarnoff) Magic of Moving Averages (Lowry) Pit Trading: Do You Have the Right Stuff? (Hoffman & Baccetti) Planetary Harmonics of Speculative Markets (Pesavento) Point & Figure Charting (Aby) Point & Figure Charting: Commodity and Stock Trading Techniques (Zieg) Profitable Grain Trading (Ainsworth) Profitable Pattern for Stock Trading (Pesavento) ‘Stock Market Trading Systems (Appel & Hitschler) ‘Stock Patterns for Day Trading (Rudd) ‘Stock Patterns for Day Trading 2 (Rudd) Study Helps in Point & Figure Techniques (Whelan) Technically Speaking (Wilkinson) Technical Trading Systems for Commodities and Stocks (Patel) The Amazing Life of Jesse Livermore (Smitten) The Professional Commodity Trader (Kroll) The Taylor Trading Technique (Taylor) The Traders (Kleinfeld) The Trading Rule That Can Make You Rich* (Dobson) Traders Guide to Technical Analysis (Hardy) Trading Secrets ofthe Inner Circle (Goodwin) Trading S&P Futures and Options (Lloyd) Understanding Bollinger Bands (Dobson) Understanding Fibonacci Numbers (Dobson) Viewpoints of a Commodity Trader (Longstreet) Wall Street Ventures & Adventures Through Fores Years (Wyckofl) Winning Market Systems (Appel) Contents Preface Foes Acknowledgments xiii list of Figures xv list of Tables xvii Introduction ‘xix SECTION ONE Background Chapter 1 Trading Time Frames 5 Intraday Trading 5 Short-term Trading 8 Longer-term Trading 10 Chapter 2 Trading Methods 13 Fundamental Methods 3 Technical Methods 4 A Combination of Methods 15 Chapter 3 Data Requirements 7 Chapter 4 The Case for Short-term Trading 19 Chapter 5 Modeling and Simulation of Trading Systems 23, Trading System Modeling 24Chapter 6 Trading System Model Simulation Modeling and Simulation Procedure Modeling and Simulation Example 28 39 eo SECTION TWO Trading with Short-term Price Patterns Chapter 7 Chapter 8 Chapter 9 Chapter 10 Chapter 11 Chapter 12 Developing a Trading System The Development Steps Daily Model Operation Example Short-term Price Patterns Automated Pattern Search The Procedure Total Short-term Trading Methodology Daily Operation Daily Price Projections Money Management Determination of Trading Capital Advanced Ways of Using Price Patterns viii 101 103 109 Trend Following 113 Pattern Clusters 116 Chapter 13 The p-Indicator 19 Background 19 p-Indicator Definition 120 Properties 121 Example 124 SECTION THREE Pattern Library Chapter 14 Pattern Library Conventions 131 CBOT T-Bonds 137 NYMEX Crude Oil 161 IMM Swiss Frane q71 CME S&P 500 Index 179 COMEX Silver 187 Inter-Market Patterns 193 Index 207 ixTRADERS PRESS, INC * Publishes books exclusively for traders and investors. + Publishes largest catalog collection of financial classiesin the U'S. * Catalogs and publishes over 650 titles forthe successful trader. TRADERS PRESS INC. ®order our 100-page catalog— hundreds of books, tapes, courses and gifts of interest to all market traders (Regular price $10) Geta FREE copy by contacting TRADERS PRESS 800-927-8222 864-298-0222 FAX 864-298-0221 Tradersprs@aol.com utp: //veww.traderspress.com Preface Making money by trading futures is probably the most difficult sone can undertake. It requires devotion, discipline and con- inuous research, In order to achieve the objective of becoming \ profitable trader, one has to make several sacrifices in his or ver personal life and must go through the painful process of sting several ideas in actual trading. This may bea very costly :deavor, but is also one that can bring financial independence. Ihere are different ways to achieve success, whether through a mechanical trading system or a fundamental approach to the sarkets. What seems to count most at the end of the day is not «» much the particular method used, but the methodology it- cit, Most traders have spent long hours in modeling and test- ing, trading systems, but few have spent the time and effort to slevelop a specific methodology and trading style. This may sound a little abstract, but years of experience have shown that 4 trading methodology is as important as the trading system itself. Of course, implementation and execution of a trading methodology require knowledge and experience that few pos- sess, as Well as a discipline that even fewer exercise. Those very tew who combine both are the long-term winners. Writing a book requires a lot of effort, but it is a very important Step in the author's life. It is a turning point, like reaching the ond of a highway and looking for a new direction to take. It also a motivation to turn to new ways and investigate new fron- tiers. That is the beauty of sharing past knowledge and experi- ence with others. MH. xiAcknowledgments !he author wishes to thank the following individuals: 1d Dobson and his staff at Traders Press, Inc., for taking a chance ona first-time writer, \werill Strasser - For providing motivation for the completion of this manuscript through his independent evaluation of the author's research work. Neil Weintraub - For his invaluable input with information on publishing a book, desperately needed by a new author. \rtemis Pulaka - For her invaluable input and assistance in preparing the manuscript. xiiiut 3 4 o4 10-1 10-2 List of Figures ‘The Market as a System The Market Model Model Simulation Modeling and Simulation Modeling Example Historical Performance Results Graph of Quarterly Returns Historical Performance of Improved Model Graph of Quarterly Returns of Improved Model The Development Steps Procedure to Generate Daily Entry Signals Pattern Example Code for the Pattern Example Shifting the Bar Labels in Pattern Example Testing Results for the Pattern Example Daily Report Generation Exact Patterns Matched Patterns Example of a Delay Pattern Automatic Search for Patterns Procedure to Generate Daily Entry Signals Daily Report Example xv 25 27 32 41 43, 45 47 48 58 63 65 66 68 69 70 74 76 78 92 93121 12-2 12-3 12-4 12-5 13-1 13-2 141 Successive Trading Signals Coincident Trading Signals Clustered Trading Signals Short-term Patterns on a Trend Historical Test Results p-Indicator Historical Performance p-Indicator Performance - Long Trades p-Indicator Performance - Short Trades Pattern Conventions xvi 110 11 12 14 15 125 126 127 135 ET List of Tables Comparison of Trading Time Frames Commonly Used Performance Statistics Examples of Targets and Stops Examples of Minimum Trading Capital Requirements xvii 20 29 105Introduction Ihis book was written with two objectives in mind: to provide hutures traders with specific trading systems, and to provide a methodology to employ these trading systems in systematic trading. Both of these elements working in synergy are required to win in futures trading. An effective trading system puts in place the prospect of profitable trading. In turn, a skilled trader uses a trading methodology to take advantage of this prospect i a way that is appropriate and is consistent with the require- ments of the trading system in use. This harmonious coopera- tion of the trader/system combination will eventually lead to consistent long-term profits. The reader will notice an absence of unnecessary illustrations of historical price charts. Only those absolutely necessary to explain the concepts presented have been included. Nowadays every trader has some means of looking at historical charts, in- cluding several daily publications and the Internet. There is no need to buy a book to do that! Furthermore, lavoid stating quali- tative trading rules such as “the trend is your friend...” etc. Most of us have heard or shared many rules that must be followed in order to be a successful trader; these sayings are often of gen- eral context and are even contradictory at times. What are rarely heard or shown are specific methodologies and systems that make money in a consistent and systematic way. The global financial system, an important part of which is the futures markets, is a very complex, dynamic process. Advances in technology have facilitated the rapid flow of funds between investment vehicles that can be located on opposite sides of the xixglobe. It seems difficult for even the most well-educated, well- trained human minds to predict the movement of capital in a way that will lead to systematic gains. Sudden reversals in the direction of market prices can make previous hard-earned gains disappear in a matter of hours, or even turn into devastating losses. Recouping the losses can be a difficult task. The systems and methods shown in this book attempt to deal with the nature of the futures markets by adopting a short-term trading approach based on historical price patterns. These pat- terns can be easily programmed in the computer and monitored ona daily basis. Positions may be placed either on the close of the day that the pattern formation is completed or at the open of the next trading day. A profit target and a stop loss are placed immediately, as soon as the position is established. The aver- age duration of a trade can vary from one to a few days, de- pending on the profit and loss objectives of the particular pat- terns employed in the trading system model. This book is divided into three sections. Section One, entitled “Background,” provides a review of basic concepts. In chapter 1 I describe the differences between the various time frames used in trading futures. In chapter 2 I review the trading methods and techniques employed by traders, and in chapter 3 discuss the data requirements of trading. Those first three chapters also provide a rationale and justification for the selection of short- term trading as the trading style most suitable for systematic trading, which is further elaborated in chapter 4. In chapter 5 1 present a comprehensive framework for modeling and simulat- ing trading systems, a step-by-step procedure to be followed by the trading system developer. This framework increases the probability that the trading system models developed will be- have the same way in real life as during the simulation procedure. xx | nally, in chapter 6, I give a specific example of how to model, unulate and improve the performance of a trading system. Ihese last two chapters of Section One provide the necessary lackground to develop consistently winning trading systems ‘ning the concepts presented in the remainder of the book. ection Two, entitled “Trading With Short-term Price Patterns,” deals with the development of short-term trading system mod- els based on historical price patterns. In chapter 7 I outline a procedure for price pattern searching based on a trial-and-error inethod. Then I present a methodology for developing short- term trading systems that are collections of short-term price patterns, and for using those trading systems to generate daily trading signals in a systematic way. In chapter 8 I classify short- ‘erm price patterns into different types, depending on their prop- erties. Chapter 9 introduces the idea of automating the search tor historical price patterns; in chapter 10 I present a general approach to short-term trading using an automated pattern search. Chapter 11 deals with money management techniques and focuses on the estimation of trading capital and risk. In Chapter 12 I discuss advanced ways of using short-term pat- terns by taking into consideration some special situations that often arise. Finally, in chapter 13 I introduce the p-Indicator, a short-term trading indicator based on historical price patterns— the first one of its kind. Section Three is a library of specific price patterns, for several commodity futures, which have been found using the automatic pattern search method I have developed. I describe the chart formation, performance characteristics, and programming logic of cach pattern listed, so it can be readily used in a trading system model. This is the first time ever that such a great number of short- term price patterns have been revealed in a single publication. xxiThe methods and techniques presented in this book are the re- sult of extensive research and development. The computer pro- grams used to discover the short-term price patterns listed in Section Three have taken many years to develop, and substan- tial resources have been devoted to that effect. I certainly hope that the reader will be motivated by this work and will proceed to develop his or her own methods and techniques of trading, using the concepts presented here as a foundation, After all, trad- ing is a personal endeavor, one that requires a combination of hard work and continuing effort. Let us keep this last comment in mind while turning the page to Section One. xxii SECTION ON BACKGROUNDBasic Concepts his section is a review of a few basic concepts useful to anyone who is involved or is planning to get involved in developing a mechanical trading system. Although very basic and seemingly trivial, these concepts play a crucial role in the development of 1 successful trading system and its correct application to real- lite trading. A lack of their understanding often results in a faulty system and in trading losses. For example, a trading system developer must understand the concepts of trading time frames, trading techniques used and information requirements that ac- company them. A trend-following system, for instance, is de- signed to follow long-term price movements, and should not be used for intraday trading purposes. Furthermore, an intraday trader must understand the need for a real-time price quote sys- tem and must pay the associated cost. As much as one may think that these are obvious conclusions, during my trading years I have come across several traders who attempted to use a trend following system for intraday trading purposes, and others who managed to trade very heavily on an intraday basis by looking at the delayed futures price updates of a television station busi- ness program. Needless to say what happened to their trading capital... The following is a list of the basic concepts that must be under- stood by a trading system developer: Trading Time Frames Trading Methods Data Requirements Systems Modeling SimulationChapter One Trading Time Frames the Trading Time Frame refers to the average duration that a irading system is designed to stay in a position, long or short. in principle, we can distinguish three different time frames: intraday, short term and longer term. What is important, from the trading system designer’s point of view, is that the model- ing and analysis process may differ, depending on the time frame selected with which to operate. Likewise, the daily trading sys- tem requirements, in terms of data needed for operation and discipline on the part of the trader, will also vary. Here I attempt to explain these differences and to cast some light on this seem- ingly trivial but often confusing subject. Intraday Trading In intraday trading, individual trades can last from a few sec- ‘onds to several hours. Usually an intraday trader establishes a “flat” position by market close and does not carry open posi- tions overnight. wIn order to develop an intraday trading system one must use historical intraday data. Although some developers make the point that intraday data looks like daily data, nothing can re- place the validity of the performance results obtained by using actual historical intraday data. The past performance of an intraday trading system must be back-tested using several years worth of data. Although histori- cal intraday data is widely available by several vendors, its in- tegrity is often questionable. A more reliable source to obtain intraday data is directly from the exchanges, or from authorized re-sellers that use the exchange “Time and Sales” records to ex- tract the data. Even if a profitable intraday trading system is successfully de- signed, its actual operation may vary significantly from simu- lated use. There are two basic reasons for this: slippage and market liquidity. Slippage is the variation between the price that a mechanical trading system indicates that an order is filled and the actual fill price. In simulated trading system performance the slippage cannot be estimated or even closely approximated because the actual trading conditions are simply not known. Although simulation packages allow for the inclusion of slip- page as a parameter in the design process, “fast” markets can result in huge deviations from expected values. This may eas- ily wipe out a streak of winning trades and leave the trading account in the red. In order to account for such conditions, the designer must perform statistical analysis based on random slip- page variable inputs—an academic researcher's task, beyond the handle of the average developer. Market liquidity, when low, can significantly alter the perfor- vance of an intraday trading system. In many instances, during ntraday position initiation, the opposite side of the trade is the ‘loor trader. Undoubtedly, when competing with the floor trader the winning chances are very small. One reason for this is that the floor broker is in a better position to know the “order flow” on the exchange and, under certain conditions, can “move” the market in a specific direction, enough to “stop-out” an intraday trader. This is especially true if the floor broker is a market maker and operates in a slow and illiquid market. This phenomenon should not be viewed as market inefficiency; it is just the way the markets work. Nonetheless, the net effect on an intraday trad- ing system’s performance can be profound. The methods and techniques used to activate buy and sell sig- nals in intraday systems, as well as to offset open positions, vary from simple and intuitive to very mathematical and algo- rithmic. The use of technical indicators in intraday trading sys- tem modeling is widespread, but their effectiveness is ques- tionable. One reason is that most of the indicators are, by de- sign, intended to “smooth” or “average” the data. That results in the loss of volatility and a lagging effect in tracking actual price moves. Chart patterns and analysis can be more effective for such applications. Intuitive methods may include conditions such as the time of the day, the day of the week, the tick volume, price change momentum, etc. ‘Commission size is very important in intraday trading. Total commissions paid accumulate fast and by the end of the trad- ing year can amount to a significant percentage of the initial trading capital. This can be studied during trading system simu lation by varying the commission variable. In my experience, it is very difficult to be a profitable intraday trader if round turncommission is higher than $12 per contract traded. Someone could ask, “If that is so difficult, why do people trade intraday?” There are a few answers to this question, depending on someone's experience of the markets and of other traders. Based on my own experience, Ihave never met an intraday trader who made money consistently for a long period of time. The explanation I attempt to give for this is that intraday trading is a full-time occupation that requires the discipline of staying long hours in front of the real-time data screen and trying to follow a trading plan without being fooled by the illusive moves of the market, At some point even the strongest deviate from their trad- ing plan. Such deviation often results in irrational moves on the part of the trader, motivated simply by greed and hope that the market will move in a certain direction... Short-term Trading In short-term trading, the trade duration may be one to several days. In essence, a short-term trader tries to “time” the direction. of the short-term price moves of a specific market. In principle this problem deals with market timing, but being successful re- quires more than that. The reason is that in the short term, mar- kets move mostly in a random fashion. The objective of a short- term trading system must be to determine the times when there is a high probability that the market will move in a specific di- rection, the long-term trend or direction of prices being unim- portant. This can only be done effectively in a systematic fash- ion; ie., there is a need for a consistently winning trading sys- tem that exhibits appropriate performance characteristics. The use of trading formulas and indicators in short-term trad- inyy is Very questionable. Formulas such as moving averages ind indicators such as Relative Strength Index and Directional Moving Index cannot be used effectively in short-term trading, Ihe time lag that these mathematical algorithms exhibit in fol- lowing short-term moves is just too long to allow their effective use. By the time a position is established the market often re- verses direction and that results in a trading loss. That is the reason why an effective method of trading short term is via his- torical price patterns, Those patterns are like the well-known chart patterns but are more exact in form and often simpler in structure, Traditional chart patterns like heads & shoulders, slouble bottoms and triangles are often subjective formations that may be more useful in establishing market turning points than short-term price direction. Unlike those traditional chart patterns, historical price patterns are more difficult to spot but are, by nature, suitable for use in short-term trading. Commission paid is rarely an issue in short-term trading. Slip- page does not affect real performance if the profit target and stop loss are kept at reasonably high levels. Furthermore, a short- term trading system can be designed (and this is often the case) to establish positions at the close of the trading day or at the open of the next day, making the presence of a real-time data feed unnecessary. And when a position is established, long or short, a profit target and a stop loss can be entered as an open position order. That leaves free time for the trader, as opposed to intraday trading, which requires the trader to watch the data screen constantly. In order to be successful in short-term trading, a system must be developed that is capable of predicting short-term price di rection. Designing such a system is not a trivial task. If historcal price patterns are used, discovering those patterns is a time- consuming process that requires extensive use of the computer. Although this is a difficult task, if done properly the result can be a consistently winning trading system that is much easier to trade than an intraday or long-term “trend-following system.” Longer-term Trading Longer-term trading, or what I often refer to as “trend” trading, is about capturing or following a trend in prices, in either the long or short direction. It is also called “trend following.” Trade duration can be weeks or even several months. The trader may choose to slowly accumulate contracts, making long-ierm trad- ing by far the method with the highest profit potential—at least theoretically. In real life there are a few basic problems that of- ten turn this stellar prospect into a disappointing endeavor. The first is that a price trend can only be identified after it has been formed! This is often called “hindsight.” In addition, the end of a trend is never known soon enough, but only when there is a considerable retracement in prices or a trend reversal! The para- doxis that in hindsight, indicators are almost perfect instruments for following trends. They are not effective, however, in estab- lishing the trend top or bottom; thus, a system developer must rely on methods such as trailing stops. The problem with trail- ing stops is that the volatility during and at the end of the trend is not known in advance. Low values of the trailing stop take the system out of the trend too soon and high values subject it toz huge loss at the end of the trend, when prices reverse direc- tion. Trend-trading systems can be designed with the use of either daily or weekly data. A real-time data feed to monitor the mar- 10 | Ison an intraday basis is not required and the trader can have snother occupation at the same time. However, even if the basic siesign problems of trend trading systems are solved, real life pplication requires extreme discipline to be exercised by the lrader. Very few traders have the patience to maintain profit- able positions for long periods of time, as one is often tempted lo pocket short-term gains as soon as they occur. As a matter of fact, [have known very few individuals who have successfully followed a trend, even partially, and profited from that! Itis by far the most difficult style of trading to implement and execute in real life. Summary Selection of the trading time frame is a very important first step in the development of a successful mechanical trading system. The selection criteria must be based on the objectives of the trader, after considering carefully the requirements imposed in terms of daily monitoring and execution. Intraday and short- term trading are recommended only if one plans to become a full-time trader. Otherwise a trend-trading system must be con- sigered. ilChapter Two Trading Methods Ihere are two basic methods that traders employ in their deci- ion process: fundamental and technical. Although the two can he combined, it is often the case that traders who use funda- mental methods rarely use technical methods and traders who ire technically oriented know little about how to use funda- mental methods correctly. In my opinion, this is possibly due to the very different educational background of the two trader types, the fundamental user being most often an economist and the technical anything but an economist. Irrespective of a trader's background and method used, success comes only if there is a deep understanding of the factors and parameters involved and their correct application in trading the markets Fundamental Methods Fundamental methods are based on the consideration, analysis and subsequent quantification of the economic, social and po- ical factors that affect demand and supply in the markets. Fundamental methods can have an application in long-term trad- ing but are ineffective in intraday and short-term trading. There 13are many reasons for this, including the limited availability of economic data in a timely fashion and the inability to predict monetary policy actions on time. Then there are cases of unex- pected events, such as wars or political decisions, which have a decisive impact on market price direction. Often fundamentals are used to make long-term investment or economic policy decisions in Fixed Income, Currency and Eq- uity Markets. Occasionally fundamentals are used to establish long-term positions in Commodity or Derivative products. The failure in late 1998 of Hedge Funds, which based trading on fundamental methods, suggests that more work needs to be done in this area. Technical Methods Technical methods are based on the use of mathematical formu- las known as studies or indicators, the analysis of chart patterns, or combinations of both. The objective is to predict future price direction only, although some chart patterns attempt to also pre- dict the magnitude of the price change. An example is the so- called “flag” chart formation. Indicators such as the Directional Moving Index have smooth- ing properties, essentially average prices, and thus are not suit- able for short-term trading due to the time lag they exhibit in tracking actual price moves. This is especially noticeable in “choppy” markets, where indicators are unable to perform. However, indicators can be excellent tools for trend following, if properly applied. Charting methods can apply to intraday and short-term trading 4 1 their effectiveness is very limited. Most traditional chart ticrns, like double tops or bottoms and flags, are very diffi- 1|l lo spot and confirmation of their formation comes very late ‘he price move, leaving very little room for profits. Historical ice patterns can be very effective in both intraday and short- m trading. There is, however, a very important assumption sacle that “history repeats itself.” Therefore, in order to use these 1s pos of patterns correctly, one must first discover them and then vudy their performance characteristics very carefully. This is ‘lhe subject of extensive work to be done by the trading system eveloper, the result of which can be very rewarding Ivchnical methods can be very valuable tools if used with ex- heme caution, Unexpected events outside the scope of techni- 1 analysis can trigger price movements that result in severe losses. It is therefore necessary to study the historical perfor- mance of any technical method considered before applying it under real trading conditions. This, of course, assumes that the wwchnical method can be fully described in a way that can be unplemented in the computer. Chapter 5 makes further refer- cnce to this subject. A Combination of Methods Several unsuccessful attempts are made, from time to time, to combine fundamental and technical methods. My opinion is that such a combination is very difficult due to the fact that the two methods deal with very different parameters and have very dif- erent objectives. Furthermore, although some economic data series can be integrated with historical price data and subse- quently applied to trading system models, their usefulness is highly questionable. One reason for this is that technical meth- 15ods and analysis already assume that prices at any given mo- ment reflect the prevailing economic conditions. Thus the use of economic data is redundant unless one expects that a sur- prise in the magnitude of an economic data release is forthcom- ing. If this is the case then the unexpected event probably will be hard to quantify and model. Furthermore, the market swings are so violent and fast when such an event occurs that trading anything becomes very difficult. In my experience, the combination of technical and fundamen- tal methods can have merit in trend trading but is to be avoided in intraday and short-term trading. Furthermore, the inclusion of fundamental factors in a trading model often causes the trader to direct attention to the daily development of such events. This type of focus may result in forming a personal opinion about market direction, which can be used to alter the decision pro- cess of the trading system. Although this topic could be the subject of an interesting debate, it is outside the scope of this book, which concentrates exclusively on systematic trading via the use of a mechanical trading system model based on techni- cal methods. Summary Fundamental or technical methods may be used for market fore- casting. Development of mechanical trading systems requires the use of technical methods that can be fully described in a way that can be implemented in the computer. Depending on the trading time frame employed, the proper technical trading method must be applied in order to obtain desirable perfor- mance. 16 Chapter Three Data Requirements Data requirements may vary significantly during both the de- velopment phase and the actual trading, depending on the type of the trading system developed and used. Intraday trading systems require the use of historical intraday ata in their performance analysis, and a real-time data feed caring actual trading operation. If the trader uses the real-time teed to collect intraday data then maintenance should be per- tormed regularly in order to find and remove spikes that are due to faulty transmissions. It is advisable to update the intraday data files on a regular basis using a reliable vendor that checks tor data integrity. If corrupted data files are used in historical simulation, it is possible that the performance results obtained will be misleading. Short-term trading systems require daily historical data during simulation and performance analysis. Daily trading requires cither real-time or end-of-day data, depending on the specific structure of the system used. If the trading system initiates po- 7sitions either at the close of today or at the open of the next day and usesa fixed profit target and stop loss, then end-of-day data is sufficient. If a short-term trading system initiates positions during the trading day, then a real-time feed may be necessary. Updating and maintaining daily data files can be done in sev- eral ways, depending on the number of futures contracts traded. There are many vendors of end-of-day data that provide a reli- able update shortly after market closing. Traders who monitor prices for just a few contracts often make manual updating. Trend trading systems can be simulated using either daily or weekly data, depending on the requirements. An end-of-day data service is more than sufficient for daily operation. Some trend traders even use the daily newspaper and manually up- date their data files. Irrespective of the methods used to update and maintain data files, careful attention should be given to any adjustments that must be regularly made to the data. The proper data Time Se- ties must be used during both the development and the daily execution of a trading system. If continuously adjusted data is used, proper adjustments must be made to past prices during the rollover of future contracts, in order to insure correct opera- tion of those systems. Most trading software packages that fa- cilitate trading system implementation and monitoring allow those adjustments to be made to the data in a relatively easy way. If the proper adjustments are not made, the trader risks compromising the operation of the trading system. 18 Chapter Four The Case for Short-term Trading lable 4-1 shows a summary of the basic concepts discussed in chapters 1 through 3. One conclusion that may be made from the comparisons on the table is that short-term trading is the only trading time frame that is better suitable to a systematic trading system approach while minimizing other requirements such as daily work needed to be performed and discipline to be exercised. I reach this conclusion by noting that short-term trad- ing requires the presence of a trading system model whereas the other trading time frames can use fundamental and intui- tive techniques. Although it is not a simple task to develop a winning trading system model, if successful in doing that, the developer puts in place the basic and necessary ingredient for tematic trading. That, of course, does not exclude the possi- bility that trading systems can be developed and used success- fully based on other trading time frames. Furthermore, a trader is free to use a combination of systems that operate in different time frames. 19section two of this book concentrates specifically on the devel- My own opinion and experience is that regardless of the tech- niques employed, consistent and long-term profitability can only ,pment of short-term trading systems which are based on his- be achieved via the systematic use of a trading system model. torical price patterns, and on the use of methodologies that fa- slitate systematic trading under actual market conditions. TIME DATA | TECHNIQUES | PROS cons FRAME | NEEDS ‘COMPETITION WITH THE INTRADAY FLOOR PATTERNS No ‘TRADER; SECONDS | REAL-TIME] INDICATORS | OVERNIGHT | HIGH TO INTUITIVE | POSITIONS | COMMISSION HOURS ‘AND. SLIPPAGE; FULL-TIME OCCUPATION FEW sHort- | REAL-TIME TRADES; NEED A TERM OR PATTERNS WELL TRADING END-OF-DAY DEFINED | MODEL ONE TOA ‘TARGETS FEW DAYS AND STOPS HIGH TREND | END-OF-DAY| PROFITS NEED OR INDICATORS | WHEN | DISCIPLINE; WEEKS | WEEKLY | FUNDAMENTAL| ADDING TO| EXITS NOT TO PosITION; | KNOWN MONTHS Low | comission} Table 4-1: Comparisons of Trading Time Frames 20 21Chapter Five Modeling and Simulation of Trading Systems The availability of high-speed personal computers at low cost hes made possible the modeling and simulation of trading sys- tems by anyone with basic programming skills. Two decades ago, this could only be done by those who had access to expen- sive mainframe machines, as well as an understanding of spe- cial computer programming skills. Today, individuals can buy software packages that provide a general platform for the imple- mentation and simulation of trading systems, using high-level programming languages specifically designed for that purpose. My own view is that all this technological progress has increased the complexity of trading and the amount of effort needed in order to become a successful system trader. Furthermore, de- spite all the advances, modeling and simulation of trading sys- tems remains a highly technical and specialized field that re- quires actual trading experience combined with knowledge of mathematics, statistics and computer programming. Lack of an 23in-depth understanding of the underlying concepts and tech- niques results in faulty trading system models, Trading System Modeling A model is a description of some system designed to predict what happens if certain actions are taken. The process of deter- mining the model is called modeling. The result of modeling is often put in a mathematical form, so that it looks like a formula, a set of equations or logical propositions, or a mixture thereof. The are at least two types of models: abstract and empirical. Abstract models are based on a priori knowledge of the actual operation of the system to be modeled. Empirical models are based on experiments and observations. The following is an example of a trading model: If the close of today is higher than the value of the moving aver- age of the last two days’ highs then buy one contract today on the close with profit target at the fill price plus one point and protective stop at the fill price minus one point. As shown in figure 5-1, the market can be thought of as a system that is comprised of all entities that participate in it, called the traders. The input to the system consists of news, information about this market and other markets, and future expectations about price levels, amongst other things. The output of the sys- tem is the price of a particular asset and has three possible lev- els: up, down or steady. That is true whether one considers the next trade, the next second, five minutes, the day of the week or the month of the year. 24 MARKET —p (TRADERS ) NEWS ie EXPECTATIONS DOWN INFORMATION STEADY FIGURE 5-1; THE MARKET ASA SYSTEM Although the output states of the system in figure 5-1 are few in number and simple in nature, what generates them is a highly non-linear and stochastic process that is essentially the result- ant of all the individual processes that characterize the deci- sions of each participating trader in the system. By simply know- ing the past history of the output states of this system, it is im- possible to come up with a mathematical model that will ap- proximate its future behavior to any satisfactory degree. It is just so complex that it is beyond the scope of mathematical sci- 25ence as we know it today. But even if a model could be con- structed, the inputs that drive it cannot be known in advance, as, to generate the output states for the next time interval. And to make things even more complicated, the market often exhibits an anticipatory behavior, meaning that the output states at a specific time may depend on expectations about the future val- ues of the inputs. This further complicates the process and makes the task of finding a model for the market a very difficult (per- haps even impossible) task. In practice, one can construct an approximate model for the market, in the sense that the model output will coincide with the actual market output on certain instances, called the “Entry Points” or “Trading Signals.” During those instances, non-li ear and random effects will, presumably, be minimized and will have a reduced effect on market behavior. This can allow sim- pler deterministic system models to be a satisfactory approxi- mation of the real non-linear and stochastic system. As input to our model, we can use the past price history of the market, a time series of prices of appropriate length that depends on the particular model used, as shown in Figure 5-2. In principle, the intention and desire of the system developer is. that the sum of all losses, or losing trades, will be smaller than the sum of all gains, or profitable trades, that are generated by the system over a long period of time. Modeling a trading system is more of an art than a science. De termining what is important to include or exclude from the model requires an in-depth knowledge of the inner workings of the underlying process. Although many trading system devel- opers treat markets as time series where the object is to make a 26 Jorecast or prediction, the result is often a trading sys! cannot be used in real-life trading. An example of this is intraday trading systems that are developed by analyzing a time series, ot historical prices based on a few minute price intervals or bars. Such time series are unrealistic to use since order placing and execution cannot take place instantaneously, as these systems »ssume when modeled and simulated. This is not a question of slippage, but rather a fault in the model design to account for vctual trading conditions. INPUT —_ MODEL (MATHEMATICAL) PRICE, HISTORY FIGURE 5.2; THE MARKET MODEL 27Trading System Model Simulation Simulation is the process of driving a model with input and observing its output. Often, when a model of a trading system is available, the object of the simulation is to use as the input a time series of historical prices and to obtain a set of trade entry and exit points, as shown Figure 5-3. The entry and exit points, labeled E, and X, respectively, can then be used in conjunction with the known input to calculate a set of performance param- eters useful in evaluating the model performance. The process of simulating the historical performance of a trading system model with input made of actual historical data is also called “back testing” or “historical testing.” The input to the trading system model can be made of tick, intraday of any period length, weekly or even monthly data. Combinations of these can also be used. As soon as the perfor- mance of the trading system model is obtained under actual historical conditions, the data can be filtered out or altered us- ing random inputs, in order to study the effect on the model performance. Time series of prices with appropriate character- istics can also be made up, using a time series generator. It is my opinion, however, that testing a trading system model ui der such hypothetical conditions is not as useful as testing it under actual historical conditions, unless the developer aims at studying the model robustness with random inputs. It is very difficult to make any conclusions about performance based on events that have never occurred, simply because it is not known. how the actual market would have behaved, or whether those hypothetical inputs actually would have been generated by the market. 28 ihe criteria for evaluating performance results obtained by the simulation are very subjective and depend on the trading ob- ivctives of the developer. One can construct many performance parameters that can be calculated using the simulation results. Those are often called “statistics.” In Table 5-1, six most com- monly used parameters are listed. Aside from the statistics listed in Table 5-1, there are other, more basic and intuitive ones that can be studied using the simula- tion results and a basic statistics software program. One example is the distribution of returns and of actual entry points. These distributions can give the developer a picture of the behavior of the model output and a feeling for its usefulness in actual trad- ing conditions. PARAMETER | DESCRIPTION N Total number of trades generated %P. Percent profitable trades Rw Ratio of average winning to loss cq Maximum consecutive losers Tp ‘Average time ina trade Dy Maximum draw down’ Table 5-1: Commonly Used Performance Statistics In simple language, as to demystify statistics, if a significant portion of the gains of a trading system model (simulated, for instance, over a period of ten years) occur within a specific, short period of time (let us say one month), then even if the perfor- mance is acceptable, the model may not be, depending again on trading objectives. This is especially important to fund man- agers who would like to demonstrate as smooth an equity line 29as possible. Another intuitive measure is to observe the distri- bution of entry points over time. This distribution must be as uniform as possible, meaning that the system operates under all market conditions, throughout the simulation period. This is especially desirable in the case of short-term trading models where the duration of a trade is on the order of a few days. INPUT MODEL OUTPUT | (MATHEMATICAL) PRICE HISTORY FIGURE 5.3: MODEL SIMULATION \s with modeling, simulation is more of an art than a science. Interpretation of the results obtained can only be made in the context of the model's intended operation. The results, in abso- lute terms, may mean very little. That is why the developer must always keep the modeling goals in perspective and must never be fooled by results that appear to be good but seem not to be justified by the model’s intended operation. Good simulation results do not always imply a good system to trade. Modeling and Simulation Procedure Trading system developers with many years of modeling and simulation experience know that the process of finding a profit- able mechanical trading system is not an easy one and has many pitfalls along the way. Here I attempt to clarify this subject and point out some problem areas. Figure 5-4 shows a block diagram of the modeling and simula- tion steps which need to be followed in order to increase the chances of developing a truly winning trading system that per- forms the way it is intended to. Modeling: Trading models can be described as abstract mathematical forms, as empirical logical propositions, or as a combination of both. Initially the developer must try to isolate the basic premise that the model operation is hased on and must implement a struc- ture of minimal complexity. This facilitates a step-by-step mod- eling process with easier checking for any modeling errors. Fur- 31thermore, if the basic idea that the model is based on generates disappointing results, increasing the complexity of the model by adding to it will not necessarily make it perform better. MopELING — |—»> IMPLEMENTATION |» siMuLATION | INTERPRETATION MODEL VERIFICATION IMPROVEMENT oF RESULTS VALIDATION FIGURE 5-4: MODELING AND SIMULATION ‘The number of variables that can be adjusted, or optimized, must be kept to an absolute minimum. As a matter of fact, one should avoid the use of any variables in a trading model, if possible. Of course, that is impossible when employing indicators or mathematical formulas such as moving averages. In that case, a safe approach is to obtain performance results for wide ranges of those parameters that are acceptable to a reasonable extent. If 32 the results have wide variations and even show negative per- tormance for a small change in parameter values, then that is an indication of over-optimization or “fitting,” and the model struc- ture must be re-evaluated. The short-term price patterns that are listed in section three do not have any variables to be optimized or varied. Irefer to them as “pure,” meaning that they are simple structures made up of daily price variations. Their modeling is very simple, and that minimizes the possibility of modeling and implementation er- rors. Model Implementation: Implementation means describing the model in a way that can be understood by the computer. In order to do that the devel- ‘oper must write computer code. This can be accomplished by either writing code based on any popular programming lan- guage or using available trading system development software packages. The latter provide a general platform for describing, and simulating trading models in a high-level language. They have a built-in database of market prices, indicator formulas, or even ready-made trading system modules. Their presentation capabilities are often stunning, providing excellent graphics and performance analysis results. Although they may sound like a bargain, these packages have several inherent limitations that are often unknown to the novice user. Therefore, one must be- come very proficient with the use of these packages before any- thing significant can be developed. Put simply, the degree of proficiency is directly proportional to the time spent using those packages. Experience has shown that it takes about two to three years of learning curve to become an efficient trading system 33developer. That may vary depending on background and pre- vious experience. Model Simulation: The object of simulation is to drive the model with historical data and generate a set of entry and exit points. This simplest form of simulation is also called “back testing.” Using the entry and exit points generated, a set of model performance param- eters can be calculated. In order for those parameters to have statistical significance, a sufficient span of historical input data must be used. In addition, the output of the model must gener- ate enough entry and exit points. It is recommended that the whole available price history be used as input data. In terms of the model output, the number of entry and exit points that is sufficient depends on the trading system time frame. Trend-fol- lowing trading models can be allowed to generate fewer trades as compared with short-term and intraday models. The suffi- cient number of output points is a debatable subject. Many (mostly novice) trading system developers take a quick look at the simulation results and decide in a few seconds whether that is a good system to trade or not. However, unac- ceptable results can also be due to errors in the model program- ming implementation or errors in the modeling itself. In any case, there are more steps to be followed in the development of a successful trading system. Model Verification: After the simulation run, the developer must verify that the M4 model operates the way it is supposed to. Due to programming limitations in the software program used or errors in the mod- cling and coding, the results can be quite different than they should. To verify the trading model results, the developer must take several batches of random data and make manual calcula- tions. Likewise, entry and exit points must be selected at ran- dom and checked manually to determine if their presence is compatible with the model logic. Verification is a time-consuming and tedious step in the devel- opment process. Its importance, however, should not be over- looked or underestimated. As a developer gets more familiar with trading system modeling and simulation, it may not be absolutely necessary to check the calculations manually. How- ever, it is recommended to always select more than one entry and exit point and investigate their validity, in spite of the sim- plicity of a trading model’s logic. Often, trivial errors can pro- duce very misleading results. Model Validation: Validation problems arise when there is a disparity between the model software operation and the way the model operates in real trading. For example, problems caused by the so-called “Temporal Aggregation” phenomenon can significantly affect the actual model performance. This particular problem arises when different events occur during the same simulation period, let us say a daily bar. Widely used modeling and simulation packages make gross assumptions when encountering such problems. An example is when both a stop and a profit target 35are hit during the same simulation bar, whether that is a five minute, a daily or a weekly period. Most software packages assume that the stop is filled first, which is a worst-case sce- nario. In order to study the effects of this phenomenon on the model performance, the developer must use intuitive measures and techniques. For instance, one can vary the magnitudes of the profit target and stop and note the corresponding perfor- mance. If the profitability of the model increases or decreases significantly then that indicates certain instability and the model logic must be carefully reconsidered. Another, more complicated way is to use second data input, with smaller time period bars, in conjunction with the model input. At least one vendor of a popular software package available in the market has realized this problem and allows for multiple time periods in a simula- tion. Nevertheless, the validation step is a very important one and there are not any generic procedures for approaching it. The developer must rely on past experience for that. Interpretation of Results: The simulation results must be studied carefully in order to decide whether the trading system model can be employed under actual trading conditions. The magnitude of the perfor- mance stalislics generated must be analyzed in the context of the model objectives and the developer trading objectives. A trend-following system, for instance, must be allowed, by na- ture, to have a higher number of consecutive losers than a short- term trading system. There are no “rigid” guidelines for this number, and the effect is the plurality of trading systems that exist in the market, many claiming to possess good performance characteristics. 36 Besides the performance statistics that are commonly generated and studied, the developer may look at several others that can be of an intuitive nature, but are often very useful. As an ex- ample I can mention the distribution of quarterly returns. This data can be obtained easily with the use of a simple statistics software package or a spreadsheet. The information presented by a simple graph like that is often very useful. In any event, interpretation of the simulation results is a very subjective pro- cess that improves as the experience of the developer increases. Model Improvement: This is a step in the process where caution must be exercised Many developers spot the worst trades generated by their trad- ing model and try to employ schemes that will help to elimi- nate them. This is no longer model improvement but more a way of trying to fit model performance to someone’s expecta- tions. Losing trades are not necessarily bad, if followed by a streak of winning trades. By adding more variables and param- eters to the model, one risks the elimination of good trades and a reduction in model performance. Ifa way to improve the model results in a severe reduction in the expected number of trades, then that must be avoided. Optimization of the model parameter values, often used in con- junction with technical indicators, can only be useful if the model generates acceptable results for a wide range of values. strongly recommend that the number of parameters that can be varied be kept to a bare minimum. The minimum number depends on the model logic, but usually more than two variable parameters ina single model can result in “over-fitting.” 37Another way to approach model improvement is to separate the logic used to generate entry points and concentrate on im- proving the logic of generating the corresponding exit points. In this way, the effect of the model improvement process on the core model structure is minimal. However, so are the chances of a major improvement! If any changes are made to the model in order to improve its performance, then the steps of simulation must be repeated, as shown in figure 5-4. Chapter Six Modeling and Simulation Example This chapter considers an example of a simple trading model. The procedure described in chapter 5 will be followed in order to illustrate its practical use. Model construction: In this step the trading model under consideration is described in simple language. The model is intended for trading T-Bond Futures contracts. The process that generates the entry points is as follows: If the close of today is higher in value than the moving average of the last two days’ highs then buy one contract on the close of the day. The process that generates the exit points is: Exit all open long positions at the close of the next day. 39Modeling: In this step we describe the model in a way that can be imple- mented in the computer. In order to do this the model logic must be described in mathematical terms. Let us consider the following parameters: C(0) = close of today H(0) = high of today; H(1) = high of yesterday; Then the model entry signal trigger is given by: If C(0) > (H(0)+H(1))/2 then establish a long position; The model exit signal trigger is simply the close of the next day following the day that the entry point is established: If system in an open position then exit at C(0). The model is shown in block diagram form in figure 6-1. The input to the model is the history of prices for the close of today (0), high of today H(0) and high of yesterday H(1). In the case of historical testing, these input prices must be available for each date in the desired historical time span. During actual model operation, and for the purpose of trading, the input prices are needed only for the current date. In either case, the model gen- erates an entry signal if the corresponding entry condition is true. If a position is in effect, then the model exits at the close of the current date. 40 INPUT MODEL. Ce F00> HOHHUNOTIENE, PRICE HISTORY, THEN ATCO) HA), H(0), C(O) FIGURE 6-1: MODELING EXAMPLE Implementation: To implement our model for back testing, we can write a pro- gram in a programming language or use a software package that allows trading system model implementation and testing. Here, by the kind permission of Omega Research Inc., we are using Easy Language. The implementation in this high-level language is extremely simple and is as follows: 41For the entry part, IF C > AVERAGE(H,2) THEN BUY TODAY ON THE CLOSE; For the exit part, EXIT TODAY ON THE CLOSE; Easy Language does everything else automatically. It has built- in formulas for the moving average; it indexes the data prop- erly, assigns the proper values to the variables and keeps track of open positions in order to know when to check for exit condi- tions. Those are just some of the many features of the software packages developed by Omega Research, Inc. that implement Easy Language. Simulation: In order to simulate the historical performance of this trading model I use continuously adjusted historical data of T-Bond futures for the period of 1/1/88 to 12/1/98. Commission is set at $18 round turn. The results are shown in Figure 6-2. The out put provides a set of parameter values that can be used to ana- lyze the model's historical performance. We may notice the low profitability of 49% and the small profit over the period of al- most 11 years of only $14,741.25. Although the profitability is below 50%, more trades are losers than winners. The model is. profitable because the ratio of average winning trade to aver- age losing trade is 1.24, meaning that on average the system makes 1.24 times more in every winning trade than what it loses in every losing trade. 42 Ferfornance Summary Equity Chart Mighlights Uiew-nodel Options 12/99 ~ All trades ‘Space bar to toggle display Total net profit gua. 48.25, Gross profit $31,074.25 Gross lost $-76,333.00 Total # of trades 405 Fercent profitable 49% Humber winning trades 199 206 Largest winning trade $1,669.50 Largest losing trade §-1,500.50 fiverage winning trade $457-66 fverage losing trade § “370.55 ] Patio aug vinsavg loss 1.24 fug trade (win & loss) $36.40 ] nex consecutive winners 7 tae consecative lovers ° tacks inimert 1 hig ethane in Towers t ] max closed-out draudown —$-9,760.75 Max intra $9,760.25 Profit factor 1.19 Max # of contracts held 1 Account Size required $12,260.75 Return on account 115% ‘Systen Uriter Plus —— nega Research, Inc Figure 6-2: Historical Performance Results Reprinted with permission of Omega Research, Inc. Figure 6-3 shows a graph of the trading system quarterly re- turns. These returns are calculated based on a $20,000 initial trading capital per contract traded, and do not include any in- terest or re-investment of profits. The graph helps in noticing 43which years have bad performance and also in spotting con- secutive quarterly losses. In the example, years ‘88, '89, 94 and ’97 show negative perfor- mance, with year '94 being the worst performer. This visual in- formation can be valuable in the study of the performance of any trading system model. It is also very easy to generate using a spreadsheet or a statistical analysis software package. Model Verification and Validation: The first priority is to verify the correct operation of the imple- mented model. Starting from the logic that generates the exit points, the developer must verify that trades last one trading day maximum; i., the model exits all trades at the close of the next day of the trade input. Looking over entry and exit dates can easily accomplish this. If a trade is found that lasts more than one day then there is something wrong with the program- ming implementation. The entry logic can be verified by randomly choosing a few en- try poinis generated by the model, and performing the calcula- tion for the condition listed in the Modeling step. In this case that task is easy, since the two highs of the last two days must be added, the result divided by two, and compared to the last day’s close. With Easy Language, and for the purpose of this simple trading system, if the developer has chosen the right function to calculate the moving average, this step is not really necessary. If the developer has chosen to write a computer program for the simulation, then this step is highly recommended. 44 QUARTERLY RETURNS GRAPH FOR THE EXAMPLE 29 x0 & # 20 20 e 3 3 10 10 é : ll eee ° z ie & 1d a] 5 & ad 2 & £ ad 20 oo ao 9 or ae ea es 0S 8 eT ee 86 YEAR Figure 6-3: Graph of Quarterly Returns Interpretation of the Results: The results obtained seem to indicate a poorly performing trad- ing system. The low profitability in conjunction with the low ratio of average win to average loss is undesirable. However, the presence of the high number of trades, 37 per year on aver- age, leaves some room for possible model improvement, and that is what will be attempted next. Furthermore, the results 45from the inspection of Figure 6-3 suggest improvement of the quarterly returns for the years with negative performance. There- fore, the developer may look at ways of reducing the trading losses while, at the same time, increasing the gains. Model Improvement: Without adding any new variables to the model, I will attempt to improve its performance by improving the exit signal part. Thus, I will abandon the exit signal used and will incorporate a profit target and protective stop loss into the model. In Easy Language that can be done in a very simple way and all the necessary checking is done by the program. A profit target of a full point and a half (48 ticks) and a protec- tive stop of a full point (32 ticks) is applied. The simulation result is shown in Figure 6-4. The profitability of the model is now above 50% and the total profit has increased significantly. However, that was done at the expense of higher drawdown. The drawdown is higher in year ’94 where the model now suf- fers greater losses. This is shown on Figure 6-5, a graph of the quarterly returns of the new model. Although years '88, ‘89 and 97 have now turned to positive performance, the loss for year ‘94 is higher. Therefore, one thing was gained at the expense of another. Please note that the best use of the graphs showing the quar- terly performance is to study individual performance and not to make relative comparisons on the magnitudes of the quar- terly returns. This is true since both graphs are calculated based on a $20,000 initial trading capital but the two models have dif- ferent account size requirements. Comparisons of return mag- 46 nitudes may require proper adjustments to be made in the ini- tial trading capital of the second system to reflect the higher «drawdown presence. Ferforsance Suanary Todel + ERAATL Chart Highlights” Ulew-nodel Options 12/99 ~ All trades Space bar to toggle display Total net profit, 963,271 Gross profit, $175,282.25 Gross loss -112,011.25 Total # of trad 228 Fercent profitable Munber winning 410 Namber Losing trades Largest vinning trade $1,808.25 Largest losing trade fiverage vinning trade 31,4054 fuerage losing trade Ratio aug uinvaug loss 1.46 fog trade (uin & Loss) §277'50 hax consecutive winners 8 Mex consecutive losers 6 fog # bars in winners 6 Aug # bers in losers 4 ax closed-out draudoun $-13,968.09 Mex intra Wtoun$-14,655.50 Provit factor 1.56 tax fof 1 Account size required $17,685.50 Return on account Systen Uriter Plus Figure 6-4: Historical Performance of Improved Model Reprinted with permission of Omega Research, Inc. 47The new simulation results also indicate that the number of trades has been cut almost in half. The reason for this is the new exit logic which causes the trade duration to be longer, an aver- age of 6 days for the winners and 4 days for the losers, as com- pared to the original model where trades last only one day. Therefore, there are now entry points that are triggered but are not taken into consideration because a long position is already in place. That is the way the testing software package used op- erates. There are other software packages that allow position accumulation, if that is the intention of the developer. (QUARTERLY RETURNS GRAPH FOR IMPROVED MODEL PERCENT QUARTERLY RETURN 88 89 90 91 92 93 94 95 96 97 98 99 YEAR Figure 6-5. Graph of Quarterly Returns of Improved Model 48 ihe developer must repeat the Verification and Validation steps tor the modified model. Then model improvement can be con- .idered again, unless the developer decides to use or abandon the model. Whether this example of a trading system model can be accepted and applied in real-life trading depends on the trading objec- tives set. Different developers may have different criteria for accepting or rejecting trading system performance. This area is as subjective as it can be. Furthermore, the trading system ex- ample was presented here for educational purposes only. Summary The basic concepts behind trading system modeling and simu- lation were illustrated with the use of an example. Develop- ment of a winning trading system that can be used under real trading conditions to result in long-term consistent profitability is a very difficult task that requires hard work and discipline. ‘Asa developer spends more time and effort towards this goal, the chances of achieving it improve accordingly. Luck seems to play no role here. The use of software packages that facilitate trading system imple- mentation and testing can significantly speed up the develop- ment process. This gain comes at the expense of the knowledge of the inner workings of these packages. In my experience, any- thing of significance that can provide a competitive advantage to the user must be developed from ground zero. In this way, the developer comes to possess an intimate understanding of his system. 49SECTION TWO TRADING WITH SHORT-TERM PRICE PATTERNSBackground The most important ability that a trading system model must have, in order to be used successfully under short-term trading conditions, is to time market moves. Historical price patterns seem to perform well in that respect. They can provide an indi- cation of the direction of prices with a high-probability success rate. Inoticed that in the late 80's, after I spent considerable time and effort testing the performance of trading models using tra- ditional technical indicators and formulas. By that time T was convinced that the only way to long-term profitability was to trade short term with a mechanical trading system. That started my search for a more effective approach to short-term trading. During the endless hours that I was spending testing trading systems then, I also noticed that historical price patterns could be grouped together to form a short-term trading system that had very appealing performance characteristics. The problem was finding a sufficient number of patterns to put together in a trading system model. That, combined with the slow processor speeds available at the time, made the task of implementing and testing such a trading model very difficult and time-con- suming. Nevertheless, during that period I used a search tech- nique based on a visual inspection of historical price charts, fol- lowed by historical testing by writing code in Basic program- ming language. The arrival of Easy Language by Omega Research, Inc. was a relief, as far as testing was concerned. However, the problem of uncovering profitable patterns from historical price data re- mained. Nevertheless, I managed to find a reasonable number 53of patterns in a period of two years, and I developed my first short-term trading system. I used that system to trade T-Bond futures contracts with very satisfactory results. In 1994, I joined the World Cup Championship of Futures Trading, which is or- ganized and run by Robbins Trading Company, a Chicago-based brokerage firm, and by the end of the year I finished in fourth place. That was a contest with real money and actual trading that lasted all year long. The good performance results of my short-term trading system motivated me to look for new ways to discover short-term price patterns. That led to the develop- ment of an automated system that can be used to search for price patterns. Later, after noticing some special properties of price patterns, I developed the p-Indicator. This technical indicator is based on price patterns and can be used to trade short term. My research and development efforts, as of the writing of this book, are still in full effect. I strongly believe that the moment one abandons the effort of developing something new and bet- ter, itis the beginning of a decline to come. Regardless of what has been achieved in the past, there will always be something better to look forward to. And always someone who is search- ing for it. 54 Chapter Seven Developing a Trading System The Development Steps In order to develop a short-term trading system based on his- torical price patterns, I first select a market to trade—for instance, the T-Bond Futures at CBOT. I obtain historical price charts, print them out and then start to work my way into discovering pat- terns. The patterns I search for have a simple form, which is based on the open, high, low and close of daily bars. I try to look as far back as ten days, but that can be hard. Up to five days of pattern length is a reasonable objective with the visual search. Figure 7-1 shows the development procedure followed. That may serve as a methodology for short-term trading system de- velopment, providing a systematic framework that keeps the developer in focus. The steps are as follows: 55Identify Pattern: The pattern identification is manual and is based on a visual inspection of the historical price charts. I concentrate on chart formations that generate a timely move of more than two full points in the price of the T-Bond Futures contract. When I spot a chart formation that seems to satisfy these requirements, I iso- late it from the rest of the chart and draw it on a blank piece of paper. Then I attempt to identify the logic for its implementa- tion Implementation: The pattern logic is implemented in Easy Language. That speeds the implementation and, subsequently, the testing process sig- nificantly. The problem with implementation is to decide which of the pattern characteristics need to be included. That is a trial and error process that results in a continuous feedback loop between the implementation and the simulation steps. There is no recipe for doing this besides the trial and error way. It is just time- consuming. Simulation: A trading system model based on a single pattern is tested on historical price data of the T-Bond Futures contract, going back to the beginning of the price history. The following values for the performance parameters are desired: Percent Profitability: > 67% 56 Number of Trades: > 20 Maximum Number of Consecutive Losers: << 4 A profit target and protective stop, each a full point in magni- tude, are used in testing. That is equal to 32 ticks of movement, or $1,000, per contract traded in the T-Bonds. These values are selected as optimum and that was the result of a historical vola- tility study aimed at maximizing short-term move size while minimizing the duration of a trade. The trade-off comes from the fact that the higher the profit target or stop, the longer the system remains in a position. At the same time, low values for the stop loss take the system out of the position too soon. High values of the stop loss also subject the system to wide equity variations. Whether a particular pattern is accepted or rejected depends upon the performance parameter values. When the number of trades appears to be small, the pattern logic is simplified and tested again. Most often that results in a degradation of the per- formance, but at times the result is better. The process is repeated with several variations of the logic and the best performance is selected, if any is acceptable. If none is acceptable, the pattern is discarded. Add to Model: Short-term price patterns with desirable performance character- istics are added to a model that is made up of individual pat- tern units. The model is implemented in Easy Language and individual patterns are allowed to interact; that is, if the model is ina long position, a pattern that generates short positions can become active, closing out that long position before a target or 57stop are hit, while also establishing a new short position. Easy Language allows that in a simple way, using global variables. atettnnde | MPLEMENT IN SIMULATE PATTERN EASY LANGUAGE PERFORMANCE suey [SIMULATE MODEL PERFORMANCE la—| app to mone IMPLEMENT ves, YES _, | sepaRaTE MODEL FOR DAILY RUN ne REMOVE FROM MODEL, DISCARD FIGURE 7-1: THE DEVELOPMENT STEPS Simulate Model Performance: Each time a new pattern is added, the resulting new trading system is tested to simulate its performance under historical 58 price conditions, as well as to examine the result of the interac- tion of the new pattern with the model made up of the previ- ously selected patterns. Besides the overall performance char- acteristics, I pay attention to the increase in the number of trades of the new model, as compared with the trades of the individual new pattern. If there is no change in the total number of trades, this may mean that the new pattern is redundant. A satisfactory increase in the total number of trades would be in the vicinity of 50% of the number of trades generated by the newly imple- mented pattern alone. The interaction ratio is defined as the to- tal number of trades of the combined pattern model divided by the sum of the individual trades generated by each pattern sepa- rately. The final result for that ratio in the first model I devel- oped was 0.7 (ie., individual patterns interacted 30% of the time), and I accept that to be a satisfactory ratio for my purposes. If the new pattern added is considered redundant or affects the interaction to the extent that either the profitability is reduced or the drawdown is increased to an unacceptable level, then that pattern is removed from the model and discarded. ‘The pattern search then continues and the steps are repeated for each new pattern found. I stop when I think that a model with satisfactory performance characteristics has been developed. The trading system model I developed back in the early 90’s had 30 patterns that could generate long entry signals and 25 that could generate short entry signals. That gave me at least one entry signal, or trade, every week, on the average. Implement Separate Model for Daily Run: Every new pattern that is found which satisfies the desired per- 59formance criteria is first added to the pattern collection that forms the short-term trading system model, and then a separate trad- ing model is implemented for just that single pattern. The ob- jective of the separate model is to assist in daily signal genera- tion and trading. Daily Model Operation When a trading system model suitable for short-term trading is available, the goal is to use it in a systematic way to generate trading signals and trading profits. As much as someone may wish that after the completion of the model development task the goal of becoming a successful trader is accomplished, noth- ing could be further than the truth! The real work starts when a good trading model is available. The object is to have a meth- odology and procedure in place that will fit the particular trad- ing style of the trader and impose a discipline that will mini- mize actions that fall outside the scope or operation of the model. In the trading model made from a collection of short-term pat- terns, some of the entry signals initiate a position at the open of the next day, but most do it at the close of the day that the pat tern formation is completed. The latter could not be avoided since experience has shown that the close of the day, as a posi- tion entry point, is very important in some pattern formations because it is a strong indication of short-term market direction and price momentum. Therefore, the intention is to generate a comprehensive daily trading plan that would list the position status of all patterns and assist the trader in placing orders for those patterns that 60 become valid on a particular day. The trader must know in ad- vance which patterns have a possibility to become active and then prepare for position entry. This implies that the conditions that could potentially make those patterns valid are available to the trader in a presentable form. The methodology for the daily operation of the trading system is shown in Figure 7-2. The steps followed are: Update Database ‘An end-of day data service is used to update the database file of the daily prices of the T-Bond futures contract. Most data ser- vices have proprietary format but provide a software package that allows converting the data to other popular formats, includ- ing ASCII, which was used here. Run All Models in Daily Mode This is done to determine the position status of each individual model representing each pattern in the short-term trading sys- tem model developed. If any of the individual models has a position status, long or short, that is noted in the output of the system. In the past, I used the “Autorun” feature of System Writer Plus, an Omega Research, Inc. product. That was very convenient, since the model logic was already written in Easy Language, which was a part of that software package. It also served as a way to check and validate the results of the daily report and to correct any programming errors that might have been present. 61Generate Daily Report for the Trader Some of the individual patterns in the model generate entry sig- nals at the open of the next day, but most do that at the close of the day that the pattern is formed. For the latter, in order for the trader to be able to act when the pattern is formed and a posi- tion must be placed, there is a need to determine in advance the progression of the formation. Because of the number of patterns involved, this must be done in a way that is easy for the trader to monitor. One way to predict the formation’s progression is to determine which patterns are candidates to give a signal at the close of the next trading day and to list the necessary conditions that must be satisfied for that purpose. I call this the “next day projection technique.” Chapter 10 describes this technique in greater detail. For each individual pattern model, additional code is written to facilitate next-day projection. That generates an output for each pattern that is a candidate to produce an entry signal, and all the outputs are combined to form a daily report file. The file is then printed out and is available to the trader as the next day’s trading plan. Mo © Prices During the trading day, the trader monitors market prices and checks the various conditions listed on the daily report. If there is a price move that invalidates a listed condition for a particu- lar pattern, the trader then crosses that pattern out and no longer considers it a candidate to produce a trading entry signal by the close of the day. As the trading day progresses, more patterns are invalidated. Usually, as the market close gets near, there is a small number of candidate patterns to produce a signal. 62 DATA VENDOR UPDATE RUN ALL MODELS tN GENERATE DAILY DATA aity Moe ee Base ENTER POSITION 7 MONITOR PRICES AND TO DECIDE IF ANEW TARGET AND STOP POSITION IS DUE FIGURE 7-2: PROCEDURE TO GENERATE DAILY ENTRY SIGNALS Position Entering: If the prices move in a way as to trigger a particular pattern then the trader inputs the appropriate order. When the order is ex- ecuted, a profit target and a stop loss order are also put in place as a “good ‘til canceled” order. Since the profit target and the stop loss are two full points apart, the order is taken at the CBOT pit on a “non-held basis.” If the movement of prices is such that 63the trader cannot determine with absolute certainty that a posi- tion is due, then that pattern is reconsidered after a market close. Ifan entry signal is triggered then, the trader enters the position in the evening session. This may result in some slippage, but in practice the long-term variations usually average out to zero. In the case that the trading model has an open position in the market, long or short, and another entry signal is generated that is of the same direction, then the new signal is neglected. How- ever, the trader may use the indication of that signal as a confir- mation of the position in place or may decide to move the origi- nal position stops to new levels, as if the new entry has been put in place. There are many ways to treat multiple signals and consecutive signals, depending on the money management ob- jectives of the trader. The possibilities are many and have yet to be explored. An Example ‘An example of a pattern formation is shown in figure 7-3. The three daily bars that form the pattern have been labeled, start- ing from 0 for the last day or today, 1 for yesterday and 2 for two days ago. The pattern logic is as follows: Buy today on the close if Close of yesterday < Low of two days ago and Low of yesterday < Low of today and Close of today > High of two days ago and Close of today > Open of today; FIGURE 7-3: PATTERN EXAMPLE The Easy Language implementation for this pattern, for both testing and daily operation, is shown in Figure 7-4. The code written has two portions. The first is the portion used to check the pattern logic conditions. If the conditions are all true then the program checks to determine whether the end of the data file, containing the T-Bond daily price data, has been reached. The end of the file is the last trading bar available. If the last bar has been reached then the system initiates a position, as well as a profit target and stop loss. It also prints out the daily report— the status of the signal. 65Directory: CLS Printed on =03/i7/98 oaagpm ENTRY SIGNAL ‘Signal Name: H&D RT LONG Developer: HARRIS ‘Notes : HD SYSTEM Last Update : 02/17/98 04:47pm Long Entry Verified : YES ALLELE ULL LONG ENTRY ANWNAANSSAANISNANANNONUNLANUOLUDY condition1=c(1) < I(2); condition2=I(1) < (0); conditiong=c(o) > ht2); IF condition: and condition2 and condition and condition THEN BEGIN PRINI(FILEL,"H&D: LONG - EXIT AT ENTRY PRICE + 32 TICKS"); BUY TODAY GVALUE1 CONTRACTS ON THE CLOSE WITH BEGIN PROFIT TARGET AT CLOSE OF TODAY + GVALUE32 POINTS; PROTECTIVE STOP AT CLOSE OF TODAY -GVALUE2 POINTS; IND; END ELSE BEGIN IF DATE = LASTCALCDATE AND C(o) < L{1) THEN BEGIN VALUEI=H(); VALUE2=INTPORTION(VALUE1); VALUEo“INTPORTION(FRACPORTION(VALUES)"32+0.5) | VALUE5=INTPORTION(VALUE4); VALUES=INTPORTION(FRACPORTION(VALUE4)*"32+0.5); PRINT(FILE1,"H&D: FLAT, CLOSE >",VALUE2:3:0,VALUE3:2:0, * CLOSE >OPEN"); PRINT(FILE:," LOW >" VALUES9:0,VALUE6:2:0); END; SyyvUaaanssgOVAMMANNNOUN AMAZE Prepared using System Writer Plus Version 2.18 by Omega Research, Inc. Figure 7-4: Code for the Pattern Example 66 The second portion of the code is used when the end of the file has been reached but not all the conditions are true. This helps to generate the daily report, in case the pattern is a candidate to generate an entry signal at the close of the next daily bar. In order to achieve this one must check to see if the actual daily bar formations are such to allow an entry signal. That can only be true if, for the bars labeled 1 and 2 in the pattern formation, their associated conditions are met by the last two bars in the data. Then we must shift the labels of the bars to be compatible with those of the code, as shown in figure 7-5. Thus, bar 1 be- comes bar 0 and bar 2 becomes bar 1. Since there is only one condition that involves the original bars 1 and 2, we check to see whether or not this is true—but only if the end of file has been reached. If the condition is true, then the values for the rest of the conditions are printed to a file. The three conditions that must be met on the next day are now properly determined. Since the ASCII T-Bond daily data files are in decimal form, proper adjustments are made to print the values out in 32nds. The testing results for this pattern are shown in figure 7-6. The testing period is from 01/01/86 to 01/01/98. A commission of $74 was used in the testing. Intigure 7-7, a daily trading report as of the close of 02/19/98 is shown. In this report, patterns that generate long positions are listed first, followed by patterns that generate short positions. Each pattern that is a candidate to generate a trading signal at the close of the next day is listed and has been assigned a name for clarity. Of course, the trading system includes additional patterns that are not candidates to generate a signal as of the close of the particular day the report is generated, none of which are listed here. Furthermore, the daily report may look quite 67different each day, since the particular patterns that are candi- dates to generate a signal constantly change. FIGURE 7-5: SHIFT OF BAR LABELS IN PATTERN EXAMPLE. The conditions shown for the next day—open, high, low and close—are part of each pattern’s logic. For example, the pattern named RECOVER, in the Long Signal part of the report, will generate a long position signal if the close of the next day in the T-Bond future contract is higher than 122 17/32 and the low 68 during the day session remains above 122 1/32. If during the trading day the low drops below 122 1/32, let us say to a price of 121 19/32, then the particular signal is crossed out and in- validated, even if the condition for the close is valid. TSTORTCAL RESTITS~ Perfarnance somnara TodeT WD TORE [IEE Equity Chart Wighlights” View-nosel Options 12799 ~ AIL trades s Total net profit 99,692.60, Seoss profit, 9151776 80 Gross loss $6,084.00 Total # of trades 22 Percent profitable 721 Nonber winning trades 6 Manber using trades Largest vinning trade $806.89 Largest losing trade fverage winning track fverage losing trade Ratie 9ug vin/ov% fog trade (win A Loss) ax consecutive 6 tax consecutive losers fog # bars in winners 4 fig Hibars in losers Wax closed-out dravdoun $-3.878.60 Max intra-day drawdown $3,510.75 Provit factar ‘59 ox Hot contracts held 1 Account size required $5,518.75 Return on account 1254 Systen Writer Plus ==— Onega Research, Ine. Figure 7-6: Testing Results for the Pattern Example Reprinted with permission of Omega Research, Inc 69uonerouas yoday Atteq Z-2 01n8ty “va vIvalds SOTO < Nao Srier> asoTo ETE > FSOIN OF 12t > MOL 9E ZI > HOLT S g & = g 62 121 > SOTO NAdO > ASOT OF tet < MOT 92 Zzt > HOT "gars sagan "AISNI Iv ‘via ‘via ‘6F OFT > SOD O Et < HOTH gr izt > ASOIOT eet < HOTH 92 eet < NAdO. NOLLISOd DAOHS V UALN OLSNOLLIGNOD NOLLISOd 'STVNDIS.LNOHS| ui ‘A MWEA0D 1 22t > MOTOE TEI < MOT of Izt > NadO 9 Zet < ¥SOTO 61 eet > NAdOT FEI< NAO 9 EEE MOT O€ Tet > NAdO SASMAATY ‘ui fc 721 < S010 6c tet < AVG. 40. MOT € Zzt > AVGAO MOT SWOLLO uw NOLLISOd DNOTY ZING OLSNOLLIGNOD NOLLISOd ‘STVNOIS SNOT = SNOUL| 00078 Szr€0'0 :JAOW WAWININ - 000°001S #4718 LOVUINOO =2e/t=Se8§ = 2€ :401S/LADUVL NOLLT WALSAS ONIGVUL GNOW-L HOI SLINSTY NOLLVTNOTVO SWIL TRL gt2086 ‘giva :MOT SHOT :NadO "SOTO ‘oreet T cer ‘ONINAAA ONIGLVIONI ozect 70 Occasionally the trading system may generate both a long and a short position. That has happened only a handful of times in the past. In such cases, no position is taken. Conversely, the more signals that become valid at the close, the better the chances may be for that position to produce a profit. Chapter 12 will discuss, in greater detail, some interesting as- pects of trading models that are made up of price patterns. Summary This chapter presented a detailed methodology that can be used to develop short-term trading systems based on historical price patterns and apply them in actual trading. Successful applica- tion of the methodology requires hard work on the part of the developer because it involves a manual search of historical price data and extensive trial-and-error testing. It also requires disci- pline on the part of the trader to monitor and execute a detailed daily trading plan. Doing both properly results in a systematic trading methodology that can be very rewarding. 7Chapter Eight Short-term Price Patterns In this book, a pattern is defined as a chart formation that is made up of consecutive price bars. The bars can have any time period length. Patterns can only be compared in the sense of the algorithm used to describe them. Those patterns that have mul- tiple occurrences that look alike visually are just special cases in the vast universe of abstract mathematical pattern forms. Therefore, if one intends to perform a systematic search for pat- terns, it may be helpful to have a basic classification that ac- counts for different types. Here I attempt to present a non-ex- haustive list of the types of patterns that one may consider in a search. Exact: These are the most common and easy-to-understand patterns. They are “exact” because every occurrence matches the other occurrences of the same pattern 100%, in the sense of an algo- rithm. These types of patterns are also easy to compare visu- ally, since their geometric form is very consistent; ie., all the occurrences of the same pattern look alike in a geometrical sense 73However, proportions are not kept to ratio—only the relative positions of price bars in the chart formation are kept constant. Figure 8-1 shows an example of two occurrences of an Exact pattern, The pattern is made up of five price bars. Note that the relative position of all bars in each of the two occurrences is the same. However, bar 3 in the first occurrence is bigger than the corresponding bar 3’ of the second occurrence. Also, the posi- tions of the open and close of bar 2 are different than those of the corresponding bar 2’. Despite these differences, the two oc- currences may look alike to the untrained eye. it ff FIGURE 8-1: EXACT PATTERNS 74 Matche: ‘These patterns are matched to a desired “matching level” ex- pressed as a percentage. Again, that is true only in the sense of the algorithm used; ie., the matching is a mathematical com- parison and occurrences of the same pattern may not look alike visually at all. The degree of matching can be selected arbitrarily, but levels above 50% are needed to produce desirable results. A matching level of 100% results in an “Exact” type pattern, by definition. Matched patterns can be thought of as sets that are formed by the union of subsets of Exact patterns. The patterns of each sub- set “match” the patterns of the other subsets at the desired match- ing level or higher. Each subset defines a distinct Exact pattern with its own number of occurrences in history. The limiting case is when each subset has only one occurrence and matches the rest at a 100% level, which is the definition of an Exact pattern. In that case the number of subsets also coincides with the num- ber of occurrences of the Exact pattern. Figure 8-2 illustrates the concept of a matched pattern by show- ing two different Exact patterns. Each of the Exact patterns be- longs to a subset and has a given number of historical occur- rences. The union of the subsets forms the matched pattern set. ‘A close inspection can show that the two patterns resemble each other greatly but are not quite the same. Proportional: These are patterns where every occurrence of the same pattern exhibits the same proportions in terms of daily ranges, expressed 75as the distance between the High and the Low of every corre- sponding bar. That is, the ratio of the first bar range to the sec- ond bar range is the same in all occurrences. The same holds for the ratio of the second bar to the third, the ratio of the third to the forth, and so on. The motivation in searching for this type of pattern is to look for geometric symmetry that is often hidden in seemingly random data. EXACT PATTERN 1 4th EXACT PATTERN 2 FIGURE 8-2: MATCHED PATTERNS 76 Delay: These are patterns that can be of any type, but order entry oc- curs with a delay. With delay patterns, instead of entering a po- sition at the close of the day that the pattern is formed or at the open of the next day, the entry is taken after a number of days, called the “delay.” The delay period can be several days long, depending on the pattern found by a search algorithm. The motivation in searching for this type of pattern comes from the fact that some otherwise profitable pattern formations, indicat- ing a position to be taken in a certain direction, suffer an imme- diate correction due to profit taking. The delay used allows a trader to filter out the correction and take a position when the market resumes direction. Figure 8-3 shows the formation of a delay pattern. The actual pattern is made up of the three bars labeled 3, 4 and 5. The rest of the bars, 2, 1 and 0 are shown with dotted lines indicating that their relative position on the chart is of no importance to the pattern formation. Position initiation, whether that is a long or short, occurs at the open or close of bar 0. Note that the delay period is the same for all occurrences of the same pattern. Spl These are patterns that are compositions of different but con- secutive patterns types, a first and a second, in the simplest case. Every occurrence of the same Split pattern has identical corre- sponding parts, whether they are of the Exact, Matched, Pro- portional or, in the case of the second pattern, Delay type. In addition, the two parts must be in the same order in the Split pattern formation of all occurrences. When comparing Split pat- 7terns, an algorithm should match each of the two parts, or two patterns, separately. FIGURE 8-3: EXAMPLE OF A DELAY PATTERN The motivation in searching for Split patterns is twofold: first, to allow for patterns that have a greater number of days in their formation, since the splitting into two parts greatly reduces the number of conditions that an algorithm must match and there- fore increases the number of occurrences; second, to increase 78 the probability of success, since the second pattern can be used as a confirmation to the first pattern’s indication for a position entry in a certain direction, long or short. In the case of a Split pattern made up of Matched pattern types, cach part of the Split pattern is matched to a desired matching level. Usually the last part of the split pattern is matched to a higher level than the first, as it is the final formation before tak- ing a position. Inter-Market: These can be patterns of any type that appear in the data of one contract, but indicate a position to be taken on another contract. For instance, a pattern can be found that, when formed in S&P 500 Index futures price chart, indicates taking a position in T- Bond futures. The results can be quite interesting when search- ing for such patterns. The motivation for this type of search is the apparent interdependence of some markets, as market play- ers keep shifting assets from one market to the other depending on prevailing economic conditions. In order to more clearly describe the different pattern types, let us make the following definitions. Let P be defined as a pattern of the following types: E: Exact M,; Matched, k = the matching level in percent Proportional, n = proportionality ratio 79Let us also define the following operators on the above pattern types: S: Split D; Delay, j= 1,23,...m the delay in bars IM: Inter-Market We can use the definitions above to describe pattern types. For instance, an exact pattern is defined as follows: P=(E} and a matched pattern as: P= {M,J, with matching level of 80% Please note that (M,oq} = {E}, by definition. A proportional pattern with 75% proportionality ratio is de- scribed as: P=(PR,} We can also use the operators defined above to describe more complicated pattern types: A split pattern is defined as follows: P=S(P,P,|, and it is formed by linking each of the two patterns, one right after the other. The patterns, P, and P,, can be of any type. 80 \ delay pattern is defined P =DIP,}, P, being a pattern of any type. An inter-market pattern is defined as: P = IM{P,}, an inter-market pattern of any type, P,. Below, we illustrate the use of pattern types and operators to describe specific pattern formations: P, = DE}, is an exact pattern with 4 days delay. P, = D,{M,,), is a pattern matched at 95% level, that has 7 days delay. P, = S(M,,£}, is a split pattern formed by linking two patterns, one matched at a 75% level and the other of the Exact type. P, = S{M,,M,}, is a split-matched pattern formed by linking two matched patterns, one at 75% and the other at 85%. P, = S(M,,.D,{PR,,}} is a split pattern formed by a matched pat- tern to the 90% level and a proportional pattern with 75% pro- portionality ratio and 3 day delay in taking a position. P, = IM{S{M,,,D,{E}}} is an inter-market split pattern, formed by linking together a 65% matched pattern and an exact pattern with a2 days delay. Iwill omit any rigorous mathematical treatment of pattern types and operators—a topic that falls outside the scope of this book but is slated to be the subject of a future publication 81Chapter Nine Automated Pattern Search In chapter seven I described a procedure to develop short-term trading system models based on historical price patterns. The search for the patterns was a trial-and-error approach, which consisted of visually inspecting historical price charts and se- lecting possible candidates. The pattern logic was then imple- mented and back tested to determine whether the selected can- didates satisfied a set of desired performance characteristics. This manual method is a very tedious and time-consuming pro- cedure, the results of which depend on the patience of the de- veloper to visually examine large volumes of historical data and guess correctly on the possible existence of patterns hid- den in it. The type of search for short-term price patterns described above raised the following questions: * How many patterns can be out there? * Are there patterns that are mathematical forms and cannot be 83discovered by a visual inspection of the charts? * What are the maximum and minimum length, in daily bars, that patterns with acceptable performance characteristics can have? * Which parameters should be included in the pattern imple- mentation? Answering these questions requires something more than a trial- and-error procedure. It requires a systematic approach to pat- tern searching. That is the motivation behind the idea of auto- mating the pattern search procedure. The objective is to try to exploit computing power (which keeps getting faster and faster, while the cost continues to drop), and to let a system designed to do just that run as long as it needs to get results. A short way of conducting an exhaustive search that uses the whole history of available data to identify patterns that possess a set of de- sired performance criteria. The idea may be easy to conceptualize but is very difficult to implement. A software program needed to be written that would perform the search, and this seemed like a monumental task at the time that the idea was conceived. Nevertheless, the goal was set in early 1994, and a working first version of the software was available by the end of 1995. The Procedure Figure 9-1 shows the procedure followed for the automatic search of short-term patterns. 84 DESIRED DESIRED NUMBER OF : DecumkeNces | PERFORMANCI PATTERN SIMULATE IDENTIFICATION EACH PATTERN, OF PATTERNS ‘ALGORITHM PERFORMANCE ‘ADDTO DATA BASE ABSTRACT FORM DISCARD, URE 9-1: AUTOMATIC SEARCH FOR PATTERNS Abstract Form of Patterns: Chapter 8 gives an introduction to pattern types. The form, or type, of the patterns to be discovered is described here in terms of their general structure and composition of parameters. This is a general description and may vary according to the developer's objective. Such a description is necessary in order for the search algorithm to be able to compare two patterns and decide whether or not they belong in the same set. There are many different ways to approach this issue, but attention should 85be given to the computational complexity involved, since the search process is calculation intensive. Pattern Identification Algorithm: This algorithm is the “heart” of the search process. It searches the data continuously and comes up with possible candidates for patterns. The input to this algorithm is the price history of the commodity considered and the desired number of occur- rences, N, of the patterns to be found: Des ed number of occurrences: >N The length of the patterns considered by the search algorithm can be of several days, or bars, long. Simulate Performance: When the search algorithm finds a pattern that satisfies the mini- mum number of occurrences requirement, it passes the charac- teristics of that pattern to a module that is designed to test his- torical performance. That module is used to determine if the pattern found also satisfies the minimum percent profitability and maximum number of consecutive losers criteria. That is: Percent Profitability: > P% Maximum Number of Consecutive Losers: 30 Percent Profitability: > 67% Maximum Number of Consecutive Losers: < 3 As the desired number of occurrences is increased, for constant percent profitability and maximum number of consecutive los- ers, the probability of finding a pattern that satisfies all the cri- teria is decreased. To increase the probability, either the percent profitability must be reduced or the maximum number of con- secutive losers increased. Different sets of protective stops and profit targets may be se- lected for the search. Examples for different futures contracts are shown in table 9-1. 87Contract | Profit Target | Protective Stop (In points) (In points) IT-Bonds 1 1 2 2 2 1 a S&P 500 5 5 8 8 | 10 10 Crude Oil 05 05 1 1 IS-Frane 1 1 15 15 Table 9-1: Examples of Targets and Stops When considering a specific futures contract, the S&P 500 for instance, several search runs may be made for different sets of input criteria to the search algorithm and profit target and stop levels. One can only determine the appropriate range of magni- tude for each of the parameters after completing a number of searches and an evaluation of the results generated. 88 Chapter Ten Total Short-term Trading Methodology ‘A complete systems trading approach must include a trading system model with good performance parameters and a meth- odology that facilitates model operation in a way that can be used easily by the trader on a daily basis. Patterns found by the automatic search procedure described in chapter 9 can be used in concert to form a short-term trading system. What is then needed is a methodology that will use those patterns in an ef- fective way to generate trading signals. The procedure followed for daily operation is shown, in block diagram form, in Figure 10-1. This routine enforces a specific mode of operation, and must be executed daily by the trader. Furthermore, although general, the procedure is systematic because it provides an op- erational framework with rigid structure that cannot be altered during operation. It directs the trader to focus on the specific trading system execution. Of course, the ultimate responsibil- ity to conform with the results lies with the will of the individual trader and is a matter of discipline that no system can forcefully impose on anyone. 89Daily Operation First the database of daily prices is updated for each futures contract considered, and then the following steps are observed. Running of Daily Projections: For each contract considered, like T-Bonds or S&P 500 Index, the patterns found by the search algorithm are stored in the data base in a form that describes each pattern’s full characteristics. Then an algorithm that runs on a daily basis takes each pattern from the data base and determines whether it has a formation up to the current date that might produce an entry signal, either by the close of the next day or by the open of the day following thenext trading day. The patterns that are candidates are grouped together in terms of their contract type and profit target and stop levels. The conditions that must be met in order to produce a signal are then calculated. Daily Report: A daily report is printed out for each contract traded and for each level of the profit target and stop. Figure 10-2 shows two examples of reports for the NYMEX Crude Oil futures. Both examples are for the Exact type patterns with zero delay. The first output is FILE cl55cl0, which denotes the contract CL (Crude Oil) with 0.5 points of profit target and 0.5 points of stop loss. The second output is FILE cl11cl0, also for Crude Oil, with one full point profit target and stop. In both cases, the patterns gen- erate signals that initiate a position at the close of the day that the patterns are formed. The date of the daily report generation is 02/18/1998. 90 Ihe daily report of each file lists first the long patterns and then the short patterns. Each pattern is identified by the date of the first day in the data file that the search algorithm noticed its occurrence. The profit target value and stop loss value are then listed—the profitability of the pattern and then the conditions that must be met on the next trading day so that the pattern generates a trading signal. Notice that the daily report lists only patterns that are candidates to generate a position as of that spe- cific report date. There are other patterns that are not candidates because their formation up to the current date does not qualify them as such. Each day's report may be quite different from the last, in terms of the specific patterns listed. For example, in file cl11cl0, the long pattern named 840323 will generate a long position at the close of the next trading day if the close of the market is above 16.45. This particular pattern’s logic involves only values of the close for the final day of its formation. Monitoring of Prices: During the trading day the prices of the markets traded are monitored and the corresponding daily reports updated by crossing out candidate patterns whose conditions to generate a signal are invalidated. For instance, in file cl55cl0 and for long pattern 840202, one condition states: OPEN < 16.60. 91DATA VENDOR ENTER POSITION AND TARGET AND STO! FIGURE 10-1 RUN DAILY PROJECTIONS FOR EACH PATTERN IN DATA BASE (GENERATE DAILY REPORT FOR THE TRADER P PROCEDURE TO GENERATE 92 MONITOR PRICES TO DECIDE IF A NE POSITION IS DUE DAILY ENTRY SIGNALS DAILY RESULTS FILE :el8set0 ‘rade on CLOSE DATE : 980218 DELAY: 0 DATE TRG SIP_PRF% CONDITIONS sHoRT #40202 0.50 050 76% LOW < 15.68 - CLOSE » 15.68 - CLOSE ‘OPEN. OPEN < 16.60 - HIGH > 16.60 840511 0.50 0.50 76% OPEN = LOW - LOW < 15.68 - CLOSE >| 1568. CLOSE < HIGH - HIGH < 16.60 840529 0.50 0.50 76% LOW < CLOSE - CLOSE < 15.68 - HIGH ~ reas DAILY RESULTS FILE: CLIICLO “rade on CLOSE DATE : 980218 DELAY: 0 DATE _TRG_STP_PRE CONDITIONS 840117 1.00 1.00 71% LOW < 15.68 - CLOSE > 15.68 - CLOSE « 1645 HIGH > 16.60, 840613 1.00 1.00 81% OPEN lo4s $40126 1.00 1.0 79% LOW 15.68 - OPEN > 15.68 - OPEN « 16.60 - CLOSE > 16.50 CLOSE < HIGH 840202 1.00 1.00 95% LOW < 15.68 - CLOSE > 15.68 - CLOSE < OPEN OPEN < 16.60 HIGH > 16.60 40511 1,00 1,00. 81% OPEN = LOW - LOW < 15.68 - CLOSE > 15.68 CLOSE < HIGH HIGH < 16.60 £40529 1.00 1,00 8% LOW < CLOSE - CLOSE « 15.68 - HIGH = 1645 Figure 10-2: Daily Report Example 93,If the market opens above that level, let us pattern is no longer a candidate and is cros at 16.65, then the ed out. By the end of the trading day, as the market nears the close, most of the patterns are invalidated and crossed out. Order Placing: Orders are placed either at the close of the day that the pattern formation is completed or at the open of the next day. On the daily report, only the conditions for patterns that trade on the close are listed. Those that trade on the open are only listed when they are validated by the system and require a position to be placed. These, if present, are simple to trade since their sta- tus is known by the close of the trading day and the position is to be placed at the open of the next day. On the contrary, pat- terns that trade on the close of the day they are formed require some extra work to be implemented. The condition of pattern 840323 in file cl11cl0 in the report shown on Figure 10-2 is the simplest kind and the order could be placed as a “Stop Close Only” order. Other orders may be more difficult to place since the conditions may be alternating from valid to invalid near the close. If that is the case, then it is best to avoid any action, wait for the market close, and run the daily report for the next day after getting the price settlements. If the report indicates that a position should have been placed at the close of the previous day then that is done at the beginning of the night session, if any. If there is no night session then the trader must wait for the open of the next day and make appropriate adjustment to the profit target and stop loss. Conceivably, the whole process of monitoring patterns and even o4 that of placing the orders could also be automated. This can be done using some popular software modeling and simulation packages available to traders, which also allow operation with a real-time data feed. When doing this, however, one must be careful in the sense of what is intended to be accomplished. Al- though such automation may sound appealing and may solve some problems related to the amount of work needed in track- ing the operation of a mechanical trading system, it is my per- sonal view that this level of automation can alienate the trader. Furthermore, full automation may result in a lack of intimate understanding of the process that generates the trades. I have noticed several times in the past that traders who use automated trading systems often ignore the signals generated. Those trad- ers spend their trading, day watching the market prices change on the screen, instead of doing any work on their system, and thus they form an opinion about future market direction. This opinion, which is often contrary to that of their system, is the major cause of ignoring the trading signals generated by their automated mechanical system. Of course, the use of full auto- mation is a debatable subject and others may have a different viewpoint. Furthermore, depending on a trader’s previous ex- perience and use of mechanical trading systems, various levels of automation may be implemented. Nevertheless, if full auto- mation is implemented, it is advisable to also run in parallel a manual check of the results, at least for a good period of time. This manual check is needed in order to verify and validate the automated system’s trade generation process. Avoiding this step may result in faulty operation due to programming errors that are not immediately noticeable, and consequently to wrong trade entries and, most probably, money losses. 95Daily Price Projections Price patterns which generate trading signals that initiate a po- sition at the close of the day that they are formed, such as the examples in figure 10-2, require special handling in order to know in advance the conditions under which they will become active. I call this the “Next Day Projection Technique.” As an illustration, let us consider the pattern example of chapter 7, for which the Easy Language code is shown in Figure 7-4. The con- ditions that must be met in order to generate a long position entry signal are: Cl) < LQ) and LQ) < L@) and (0) > O0) and C0) > H(2) where the index 0 denotes today, 1 yesterday and 2 the day be- fore yesterday. The next day projection technique is based on a forward shift of the day indexes. That is, today becomes tomor- row, yesterday becomes today and the day before yesterday becomes yesterday. No indexes are used for tomorrow, i.e. L(0) becomes just L, C(0) just C and O(0) just O. The shift results in the following conditions: (0) < L@) and L@)0 and c > HQ Notice now that as of the close of a specific day, any of the C(0), L(0), H(1) and L(1) have a specific numeric value. The condi- 96 tions that involve a quantity with no index are those that must be satisfied by the next day price ranges, and the conditions that involve only indexes are those that must be true as of the close of the last day in order for that pattern to be a candidate for a signal. In the example above, after the shift in indexes, the following condition involves only indexes: C0) < La) and this must be a true condition in order to proceed further. If true, then the following are the conditions that involve no in- dexes and must be met on the following day: L >) and Cc >O and c > HQ) The above set of conditions for the values of the low and close of the next day, Land C respectively, must be monitored through- out the trading day to check if any of them becomes false. If, for instance, the low of the next trading day falls below the low of today, L(0), then the pattern cannot generate a trading signal, even if the other conditions are met. The same is true if the mar- ket close is either lower than the open or lower than yesterday's high, H(1). In order to present a more general approach to the next day projection technique, let us consider a price pattern that is formed by n consecutive price bars and has k conditions for generating a trading signal at the close of the nth bar. The k con- ditions involve only the open, high, low and close of any of the n price bars. The next day projection technique procedure is as follows. 97Step 1: Label all bars starting from 0, for the last, up to n-l Step 2: The open, high, low and close for the j bar are denoted as follows, in all k conditions: open = Of) high =H) low =L() close =C(j)_, forj=0,1,2,....n-1 Step 3: Derive the new conditions by subtracting 1 from all the indexes found in all k conditions. That is: Of) > Of-1) HG) > HG-1) LG) > L§-1) CG) > CG), forall j 1,2,n-1 Notice that the index “-1” is simply a no-index and is denoted as follows: Step 4: After the index shift, there are again k conditions that are separated into two categories: a) conditions that do not contain 0,H,L,C, ie. do not involve any no-index quantity and 98 b) conditions that contain any of the O,H,L,C, ie. involve no-index quantities. Step 5: Check to see if all of the conditions that do not involve any no-indexes are true. If any untrue then quit. Ifall true then move to step 6. Step 6: Solve for the ranges of O,H,L,C that will make all conditions that contain them true. The last step involves simultaneous solution of a set of condi- tions. In their simplest form, as in our example, these condi- tions may be linear inequalities that involve the quantities O,H,L and C. In that case, the solution is simple and so is the code that must be written. More complicated conditions may require some mathematical manipulations to be made. There are many ways to handle this, ranging from manual solutions to the use of sym- bolic mathematics manipulation packages. When a trading sys- tem model is used that is made up of a large number of price patterns then a method to automatically solve the conditions of step 6 is necessary. The procedure to generate the daily entry signals shown in Figure 10-2 of this chapter utilizes a simple symbolic mathematical manipulation algorithm that was espe- cially written for that purpose. 99Chapter Eleven Money Management Money Managements, all by itself, a very important and often misunderstood subject. Many traders overlook the basics of Money Management that deal with the determination of the amount of capital required to trade a system, which is the most fundamental and useful application of the subject. Instead they make attempts to figure out fancy methods and techniques for varying the number of contracts traded, either in opening or exiting a position. Their intention is to maximize gains. The com- mon mistake is that these methods are rarely treated correctly as an integral part of the model operation logic during the de- sign and simulation phase. One reason for this is the complex- ity of the code that must be written for the proper implementa- tion of such methods. The result from this oversight is that the expected drawdown is unknown, and there is a danger that the initial trading capital will not be sufficient to cover it. The discussivn that follows gives attention to the quantification of risk as it relates to determining the capital required for trad- ing a mechanical trading system. In that respect, one can distin guish two different types of risk: constant and variable. 101Constant risk means assuming the same magnitude of risk when- ever initiating a new position. This is the way that most short- term trading systems are used. The risk may be determined by the stop loss per trade and the initial trading capital. For in- stance, if the stop loss per contract traded is $1,000 and the ini- tial trading capital per contract traded is $50,000, then the risk is 0.02. Thus, it would take 50 losing trades to completely wipe out the initial trading capital, not including commissions paid. Often the maximum risk is stated as a percentage of the trading capital and experts will recommend that it must be less than 0.5%, or 1% or 2%, etc. Another approach involves using the magnitude of the maximum drawdown that will be generated by the trading system during actual operation. However, this number cannot be known in advance but only its simulated es- timate, based on historical testing results. Variable risk per trade implies an adjustment to either the stop loss magnitude or to the number of contracts traded each time per given trading capital, or both. The adjustment can be made according to some algorithmic procedure which is a part of the trading system model, and must be considered as such in the historical testing. For instance, the input to the algorithm can be the market volatility, the rate of real equity increase or decrease, etc. Estimating the risk parameters of such trading systems is not a simple task. There is also no guarantee that these types of systems will produce better results than systems that use a con- stant number of traded contracts and simple stop loss measures. Experience has shown that increasing the complexity of a trad- ing system, whether in the entry or the exit part, often does not better its performance, and can even result in degradation of the performance. 102 Determination of Trading Capital This is probably one of the most important steps in the correct application of trading system models, and therefore cannot be omitted from any discussion of the subject. Underestimation of the capital required to trade a mechanical system may result in losses and eventual failure. That can happen even if the trading system is profitable, due to what is often called “over-trading.” In practice, it is advisable to select a constant risk level and cal- culate the required minimum trading capital per contract ac- cording to the formula: M=S/R qd) where: M; the trading capital requirement per contract traded, in $. R: the risk level per trade, in decimals ranging from 0 to 1. S: the stop loss per trade and per contract traded, in $. For instance, with a risk level of 0.02, or 2%, and a stop loss of $2,000 per contract traded, the minimum capital requirement is $100,000. If the trader decides to trade one contract per $50,000 of capital, then for the same stop loss value of $2,000 per con- tract the effective risk is 4%, as can be calculated from equation (1) above. Equation (1) can apply to trading systems where the stop loss is 103known in advance and remains at a constant level, such as the short-term trading system based on price patterns discussed in chapter 10. The drawback of this method for estimating the mini- mum trading capital requirement is that a losing trade does not always generate a dollar loss equal to the stop loss level, since adverse market conditions can result in higher losses. This is especially true when there are “fast” market conditions or illiq- uid markets, or when the stop order is filled at a market gap opening. However, in the case of short-term trading systems that generate a sufficiently large number of market entry points, or trades, the longer term effect from these adverse market condi- tions tends to be counterbalanced by favorable conditions that generate higher profits than the selected profit target. Imust make note here that the common practice of most traders is to first select their trading capital size and then calculate the corresponding risk per trade. Then they trade anyway, irrespec- tively of the level of risk calculated! This is contrary to the method presented above, where the risk level is selected first, and is of much greater importance when a decision must be made whether or not to use a trading system model in actual trading. Table 11-1 shows a list of examples for the minimum trading capital requirement per contract traded, for various levels of stop loss and risk, assuming one futures contract per trade and no contract accumulation. The next step in determining the minimum trading capital re- quirement is to look at the historical performance analysis re- sults for the trading system model under consideration and to note the maximum drawdown. 104 RISK | STOPLOSS | MIN. CAPITAL (%) (8) ) 1,000 100,000 1 2,000 200,000 5,000 500,000 1,000 50,000 2 2,000 100,000 5,000 250,000 Table 11-1: Examples of Trading Capital Requirements Let M, = the trading capital requirement per contract, in $. D, = maximum expected drawdown, in $ f =a safety factor multiplier, ex. 1.75, 2, 2.25, 3,..., etc. MG = the margin requirement per contract, in $ Then M,=MG+fxD, Q) For example, if the maximum drawdown is $15,000 and the re- quired margin $3,000, then for f set to one the minimum trading capital per contract traded is $18,000. If the safety factor Fis set to 2, the new figure is $33,000. 105The presence of the safety factor f accounts for a possible future increase in the drawdown amount calculated by the historical testing. Experience has shown that trading systems tend to gen- erate higher drawdowns during actual trading; therefore, a num- ber of at least 2 is recommended for f. Finally, the determination of M, the minimum trading capital requirement per contract, comes from comparing the results of equations (1) and (2) and selecting the largest figure: M=max{M,,M,} (6) As an example of the use of equation (3), let us consider a situ- ation with the following parameters: Risk: R = 0.020r2% Stop: S = $1,000 Drawdown: D, = $20,000 Margin: MG = $3,000 Safety Factor: f = 3 the above dollar values are per contract traded. Then, from equation (1): M., = $1,000/0.02 = $50,000 and from equation (2): 106 3,000 + 3 x $20,000 = $63,000 Finally, from equation (3): M = max{$50,000,$63,000} = $63,000 Now, solving equation (1) for the risk R, we get: R= S/M =$1,000/$63,000 = 0.0159 or 1.59% Due to the new capital requirement, the actual risk taken per trade has dropped to 1.59% from 2%, because the initial trading capital per contract was increased from $50,000 to $63,000, us- ing the procedure just outlined. Of course, the final number is decisively influenced by the se- lection of the magnitude of the factor f, in equation (2). Increas- ing the value of f results in increased capital requirements and reduced return on capital, but also reduced risk, in terms of expected equity drawdown and volatility. Decreasing f results in reduced capital requirements and increased return on capi- tal, but higher actual equity drawdown and volatility. There seems to be no procedure to determine f a priori, because its optimum value depends on future results that simply cannot be known in advance. The trading system developer and trader must decide based on past experience and future expectations. 107Chapter Twelve Advanced Ways of Using Price Patterns In order to discuss some advanced ways of using trading sys- tems that are collections of a large number of short-term price patterns, the following definitions are made: Trading signals that occur one after the other in time are called successive. Every signal exits its position with the designated profit target or protective loss before the next signal arrives. The time period between successive occurrences must be short and is usually less than 15 trading periods or bars. Figure 12-1 shows an example of three successive trading sig- nals. The first signal initiates a position as “ENTRY 1” and exits that position as “EXIT 1.” The second trading signal follows after the first has exited and is labeled as “ENTRY 2,” with cor- responding exit as “EXIT 2.” The third signal appears only after the second signal has exited. 109fie | ENTRY 3 { ENTRY 2 t ENTRY | 7 FIGURE 12-1: SUCCESSIVE TRADING SIGNALS. Trading signals that occur while another trading signal has an open position in place are called coincident. The trading signal direction must be the same; ie., a long signal occurs while an- other long signal is in effect. Occurrence of the second signal must take place at least one bar after the first signal has initiated a position. Figure 12-2 shows an example of coincident trading signals. The first signal is labeled “ENTRY 1” and is in an open position while the second signal, labeled “ENTRY 2,” appears. The third 110 signal, “ENTRY 3,” comes after the first signal has exited with “EXIT 1” but while the second signal is still in an open position. Therefore, signals two and three are coincident but signals one and three are successive. ENTRY 3 ENTRY 2 4 ENTRY 1 FIGURE 12-2: COINCIDENT TRADING SIGNALS. Trading signals that occur at exactly the same time and all indi- cate taking a position in the same direction, although signals resulting from different patterns may have different profit tar- 11gets and stops, are called clustered. Figure 12-3 shows an example of two clustered trading signals, labeled as “ENTRY 1” and “ENTRY 2.” Although the two sig- nals exit at different points on the chart, they are clustered be- cause they occur at exactly the same time. Signals one and three are successive because signal three occurs after signal one has exited its position while signals two and three are coincident as signal two is in an open position when signal three arrives. The examples demonstrate that successive, coincident and clus- tered trading signals may be formed by any numbers and in any sequence, thus providing many interesting opportunities in trading. ENTRY 3 ENTRY 2 ENTRY 1 FIGURE 12-3: CLUSTERED TRADING SIGNALS 112 Trend Following When short-term patterns generate successive trading signals in the same trading direction, either long or short, it is often possible to achieve partial trend following. However, this is not something that can be achieved by design, but rather a phenom- enon that happens occasionally and only by chance. Ifa trading system model includes a sufficiently large number of short-term patterns, there is a good probability that, during market trends, successive entry signals will allow the trader to capture a good portion of the price trend. In that case, the chances for achieving, trend following are increased. When successive entry signals are generated by the short term trading system model, the net effect is to follow the price direc- tion. However, some short-term patterns often result in losing trades, even if their trading signal direction is in the same direc- tion with the price trend. This happens since short-term pat- terns tend to ignore the existence of a trend. Nevertheless, a well- designed short-term trading system will cause enough winning, trades along the direction of the trend to allow the trader to cap- ture a good portion of it as a net result. Figures 12-4 and 12-5 show the results of a historical test where successive long position signals are generated when trading T- Bond Futures using a short-term trading system model made up of a collection of patterns, long and short. The period of the test shown is from 3/1/85 to 7/1/85 and the data used is con- tinuously adjusted. Out of the thirteen long trading signals ob- tained, two are losers and eleven are winners. The total profit of about $9,000 corresponds to a nine full point move per contract in the T-Bonds. Considering then the fact that the magnitude of 113the trend, as shown in figure 12-4, is roughly thirteen full points, from the bottom formed in March 85 to the peak in June ’85, the trading system is capable of capturing 70% of that in profits. In effect then, the trading system, although designed for use in short-term trading, manages to perform as a trend-capturing system in this particular case. When a large collection of pat- terns is available, capturing a good portion of a price trend is a possibility. Figure 12-4: Successive Trading Signals on a Trend Reprinted with permission of Omega Research, Inc. 114 Equity Chart Highlights 12799 - al Space bar to toggle display ] rotar vet provi 33, 130-5 Grose profit $11,150.56 0 0 Gross loss 13. Fercent profitable ote amber vinning trades 11 Manber losing trades 2 Largest winning trade 1.57.25 Largest losing t $1,014.00 fucrage winning trade $1,014.41 Average loving t Ratio aug uinvayy Loss 100. Aug trade Cain & loss) tax consecut ve winners Nex consecutive losers flog # bars In winners. ug ¥ bers in losers, fax closed-out Provit. factor fox intra-day draudoun Nox #'of contracts held Retarn on account Figure 12-5: Historical Test Results Reprinted with permission of Omega Research, Inc. In the case of coincident entry signals, the trader may choose to move the open position profit target and protective stop to new levels. That is, if the trading system has an open long position with a profit target and protective stop already in place, and a newly generated trading signal indicates that a new long posi- tion should be taken, then the trader may decide to ignore the new signal, while moving the profit target and protective stop to a new position, as if the new trading signal was being ex- 115ecuted. If enough coincident trading signals are generated, the net effect in moving the profit target is to follow a price trend Moving the protective stop acts in a similar way with trailing stops used in trend-following systems. If a trading system model allows for variable risk, successive and coincident trading signals may be used to accumulate con- tracts by adding to existing positions. This must be done only after a careful consideration and estimation of the risk taken in relation to the trading capital. Modeling and simulating short-term trading system models that are formed by a collection of short-term price patterns, allow- ing successive and/or coincident trading signals to be used ei- ther for adjusting the profit targets and stops, accumulating po- sitions, or both, is a very challenging task. Estimating the performance characteristics of variable risk trad- ing systems can be difficult due to the complexity of the code that needs to be written. Traders should carefully study those systems and should assume worst-case conditions when esti- mating the capital requirement. Under-capitalization of a trad- ing account may result in liquidation of positions and in losses, regardless of how profitable the mechanical trading system is. Pattern Clusters T have observed that when clusters of trading signals appear, the chance that a profit will be made is better than with a single signal. The trader may treat clusters as a single trading signal 116 and may maintain the same risk level, or may decide to increase the risk level by increasing the number of contracts traded Again, as with coincident trading signals, this may result in vari- able risk, so the model performance parameters must be care- fully studied during simulation. A simpler way to use clusters is to consider two different levels of risk, the first being the risk with which the model is designed to operate, called “full” risk, and a second which is a fraction of the first, called “reduced.” The trader may decide to use the reduced risk when placing positions resulting from single or isolated trading signals, but may use the full risk when clusters occur. In this way, the worst-case risk is the one used to calcu- late the minimum trading capital requirements, and when the reduced risk is taken the trading system model operates in a ‘less risky” mode. However, this trade-off may result in reduced actual profits as compared to trading at full risk, unless enough cluster signals are present. Summary When using short-term patterns, a few special cases arise which provide interesting opportunities during trading, such as trend following and contract accumulation. Alternative ways of using short-term price patterns are many, and their study and analysis can be a very challenging task. In addition to the patterns presented in this chapter, there are many that await exploration as the focus of further research on this subject. 117Chapter Thirteen The p-Indicator Background When traders and market analysts refer to indicators, names such as Relative Strength Index (RSI), Directional Movement Index (DMD, and Commodity Channel Index (CCI) immediately come to mind. Although indicators are used in an attempt to forecast future price direction, most have very limited predicting power. The reason for this is the time lag that results from the data- smoothing properties that most of them have built in. In the case of short-term trading, i.e., taking positions in the market that last from one to a few days, most indicators are not effec tive because they cannot predict short-term market direction. Nevertheless, the concept of looking at a quantity that ranges from 0 to 100, for instance, and generating trading signals based on that, is a very appealing one. 119p-Indicator Definition The p-Indicator is a new concept in defining and constructing an indicator. It is based on the use of short-term price patterns that are created by the market. Since it is a short-term indicator, only the last few days of a particular market are taken into ac- count in the selection of the length of the patterns considered by the indicator. However, the whole price history of a particular market is considered when evaluating the past performance of those patterns. The theoretical range of values for the p-Indica- tor is from 0 to 100, but values between 30 and 70 are normal. The value of the indicator is calculated as a ratio where the nu- merator is the sum of the percent profitability of the patterns considered in it (each weighted by its corresponding number of occurrences in the data history) and the denominator the sum of all the pattern occurrences. Thus, the p-Indicator can be thought of as having two parts: one to be used for taking long positions and the other for short positions. The calculation of each of the two parts is identical, but in the long part of the indicator the historical percent profitability of the patterns considered is cal- culated based on taking long positions, and in the short part it is based on taking short positions. The p-Indicator is then defined as follows: Let 1,...N to be the number of pattern forms considered, P,, the percent profitability of each pattern for a long position and P,, for short, T, the number of “trades,” or occurrences of the 120 pattern in the price history, PI,, the p-Indicator value for long position taking and PL, for short. Then N N DPuTi PsiTi i=l Pie PIs= N N rT XTi i=l i=l p-Indicator Properties Construction: The p-Indicator is like a weighted probability function. We can think of the historical profitability of a pattern as the probabil- ity that the next occurrence of that pattern has in giving a profit- able signal. However, the more occurrences that the pattern has, the more “weight” its historical profitability gets in the p-Indi- cator formula. To illustrate that with an example, let us assume the p-Indicator is made up of only three patterns. For every pat- tern the corresponding historical profitability, P,,, is calculated for taking a long position. A specific profit target and protective stop are assumed in the calculations. The corresponding num- ber of pattern occurrences is T,, Let us further assume that the following hypothetical results are obtained: P, = 60% T, = 150 1215 5% 250 The long portion of the p-Indicator is then: 60 x 160 +75 x 75 + 45 x 250 PI, = 54.47% 160 +75 + 250 Thus, the new weighted profitability is 54.47 %, and that is the value of the p-Indicator for long position taking. Although P, and P, are 60% and 75% respectively, P,, which is only 45%, greatly influences the final outcome because it is accompanied by a much higher number of occurrences. Practical Use: When the value of PI, is closer to 100, it indicates a better chance when taking a long position in the corresponding market and when the PI, value is closer to 100, a better chance for short po- sitions. In practice, numbers from 60 to 70 give good results. Thus, in short-term trading we can use the values of the p-Indi- cator as follows: Long Position if Pl, >a, a>50 Short Position if Pl, >b, b>50 Inaddition, both long and short parts can be considered together as follows: Long Position if PI,>aand Pl,50 122 Short Position if PI, a, b<50 Calculation: Determining the values for the p-Indicator is a calculation-in- tensive process. This is not very important when considering daily data, but real-time calculations, if desired, can present a problem. Currently, 10 different pattern forms are considered, cach for two different sets of targets and stops. For the resulting 20 patterns, the close of the day and the open of the next day is applied as the position entry, thus leaving a total of 40 patterns to be calculated. In the case of daily data, the calculation of the p-Indicator may take from several seconds to a few minutes, depending on processor speed. Characteristics: It must be noted that when calculating the past performance of each pattern to be included in the p-Indicator formula, a profit target and a stop loss of specific magnitude are taken into con sideration. Thus, the p-Indicator has the concept of profit target and stop loss incorporated into its definition. As far as Ican tell, this is the only technical indicator with such a property that is found in the literature. Its usefulness results from the fact that conventional indicators generate entry signals for a particular market, long or short, but do not provide any indication what- soever to the trader as to when to take profits or losses. With the p-Indicator, the profit or loss is known in advance, as soon as the signal is taken. In addition, at any given moment, the p- Indicator considers the whole price history in its calculations. This is also the first time, as far as I know, that a trading indica- tor has been developed with this inherent property. 123Example An example of the performance of the p-Indicator is presented here. The following trading system model is considered: Long Entry Signal: If PI, >a then buy at the open of tomorrow with 1. ry 1 profit target at open of tomorrow +1 point protective stop at open of tomorrow - 1 point Short Entry Signal: If PI, >b then sell at the open of tomorrow with profit target at open of tomorrow - 1 point protective stop at open of tomorrow + 1 point Daily data for the day session of the CBOT T-Bond Futures are used from 1/18/90 to 9/21/98. The values of a and b, in the long and short entry signals, are selected to be 60 and 54, re- spectively. Commission is set to $18 round turn The value of the p-Indicator is calculated on a daily basis, with the first day of calculation being the first day of testing, 1/18/ 90. The historical data for the T-Bond futures contract start at 2/ 15/83 and are used in the p-Indicator calculations, since the whole price history is required. Figure 13-1 shows the result of the historical test. For this set of values for a and b, the trading system is 67% profitable and has generated 108 trades, which amounts to about one trade per month. The average number of bars for winners is one day and for losers, two days. The maxi- mum intra-day drawdown is $4,586.25 and the return on an ac- count of $7,586.25 is 442%. 124 Figures 13-2 and 13-3 show the historical test results for long ind short trades, respectively. The number of long trades is 52 ind that of the short 51, meaning that the trading system oper- ates in both long and short directions with the same trade fre- quency. The profitability of 69% for the long portion and 66% tor the short is quite satisfactory, as are the drawdown and the return on account. TSTORTCRL RESULTS Performance Summary Equity Chart Highlights iew-nodel Options 12/95 ~ All trades Space bar to toggle display $33,582.25 Total vet profit $65,146.25 Gross loss Gross profit, $-31,594.00 Total # of trades Number vinning tredes Percent profitable funder losing trades Largest losing $1,361.75 fuverage Losing § “957.39 fog trade (vin & loss) $325.75 fax consecutive winners fog # bars in winners fox consecutive losers 4 fug bars in losers 2 fax closed-out drawdown — $-4,242.50 Max intea-day dravdoun — §-4,586.25 Profit factor 2.06 tax 4 of contracts held 1 Account size required $7,586.25 Return on account 42 Systen Uriter Flas ——— Onega Research, Inc, ——— Copyright 1989 = Figure 13-1: p-Indicator Historical Performance Reprinted with permission of Omega Research, Inc. 125Chart Highlights ete quity "Chart Highlight iew-nodel options i Ms Beta, sting otal wot profit 5 38 Space bar to toggle displ arate prot SESE cross tose $15,060.50 Total wet profit $10,251.30 total a of tredes st tereent 66 toes protit, 853,527.00 Gross toss $15,205.50 Rinter'sfantag Srades BR % aren tec ria iy Largest winnog trade $1,450.25. Langest losing trade 91,018.00 thnber vinning.tredes 56 ander losing trades re fucrage inning trade, § oolet Largest vinming trade $1,325.75 Largest losing trade $21,361.25 Max consecutive winners - fveroge inning trade" S343-61 hucroge Losing trade Uaeeleg hy SUS 3 tio ag ving toss fig trade tein ess) eee gn Poa tex consecutive winners 7 ox consecutive losers Perit Hoeven Rey Wotteteacts neta 2 fg tare in utters 2 Recount site required Bekutn’on secon bs me 103 Mx closed-out draudoun $3,054.60. ax intraday dreudoun I] Frorit tector 2.46 tax ¥ of contracts held Account size required 66,147.78 Return ov acsoent Systen Weiter Plus —— dncqa Hescarch, Inc. —— Gy : Figure 13-3: p-Indicator Performance~-Short Trades p-Indicator Performance - Long Trades Reprinted with permission of Omega Research, Inc. Reprinted with permission of Omega Research, Inc. 126 127SECTION THREE PATTERN LIBRARYChapter 14 Pattern Library Conventions In this chapter, I provide guidelines for the use of the library of historical price patterns. These patterns have been found using the automatic pattern search procedure described in chapter 9 Each pattern is defined by the following parameters. Contract traded: This is the futures contract for which the pattern applies. Dis- tinction is made for contracts that trade during the day session only and those that include prices from a night session. Traders should exercise caution in applying the proper contract prices if the pattern is employed in actual trading. Type: This is the pattern type, as defined in chapter 8. In this pattern library only Exact, Matched, Delay and Inter-Market patterns are shown. 131Position: This is the market position, long or short, that the pattern is to be used for. In the case of Inter-Market patterns, the contract that must be used is also noted, since those patterns are formed in one contract but used to take a position on another. Profit Target: The profit target, given in full points, that must be used for trad- ing with the specific pattern. Next to the points, I list the equiva- lent dollar amount in order to avoid any errors. Protective Stop: The stop loss, listed in full points, that must be used in trading. Next to the points, Ilist the equivalent dollar amount in order to avoid any errors. Order Entry: This can be either the close of the day that the pattern formation is completed or the open of the following day. Order delay: When a pattern is of the Delay type, the delay in daily bars is listed. That is the number of days that the position entry must be delayed. If, for instance, the delay is 2 and the pattern ini- tiates a position at the open of the next day, then the position must be placed at the open of the day that follows the 2-day delay; that is, at the open of the third day after the pattern for- 132 mation is completed. If the pattern initiates a position at the close of the day that it is formed and the delay is 3 days then the position must be placed after 3 days. Logic: Figure 14-1 shows a pattern that is formed by 5 chart bars. Every bar is labeled above its High, starting from 0 for the last, which is also the bar with which the pattern formation is completed. For each bar, j, the following notation is used: High of the j bar Low of the j bar j) = Close of the j bar O(j) = Open of the j bar, 1,2,3,..5 ‘The arrows that appear on each bar denote the quantities that appear in the pattern logic. For the graphic example on Figure 14-1, the following quantities are to be included in the logic of the pattern: Bar 0: H(0), C(O) 1:C(1) 2: H@), L(2),C(2) 3: O(3), C(3) 4: 0(4), H(4), L(4) The logic of the pattern can include only the quantities labeled by the arrows. The remaining quantities are not used, and there- fore their corresponding actual position in the pattern forma- tion can be anywhere and is not significant. 133For each pattern listed in the library, the historical testing re- sults are shown on the opposite page. All historical performance results have been obtained for the time period indicated and assume one contract per trade. All commissions are set to $18 per round turn contract, unless otherwise noted. 134 FIGURE 14-1: PATTERN CONVENTIONS, 135T-BOND PATTERNS 137T-BOND PATTERNS T-BOND PATTERNS Day Session Only Day Session Only Pattern Code: UA830406-1 Type: Exact Pattern Code: UA 830406-1 Position: Long Profit Target: 1 point = $1,000 Protective Stop: 1 point = $1,000 Historical Performance Results Order Entry: Close, C(0) Period 2/15/83 - 1/7/99 Order delay: 0 Logic: L(1) > H(2) and H(1) > L(Q) and H(3) > L(2) and LQ) > L(3) and L(0) > L(1) and H(0) > H(1) and (3) > L(2) Chart Highlights View-nodel Options 1299 = All trades Spece bar to toggle display | $10,660.75, $20,565.75. Gross toss $-10,308.00 6m 31 Percent profit 21 amber Tosing $1,294.50 $-1-143.00 $3903? §-1,030 50 Aatio aug winsaug loss 097 $343.50 fax consecutive winners & Max consecutive losers 2 fog a bars in winners. 4 fg # bars in losers 2 fax closed-out dravdoun $2,157.00 Max intre-tey dravdoyn —$-2,790.75, |] Prorit factor 203 fax # of contracts held 1 Account size required $5,790.25 Return on account. ety Systen ueiter Plus — Onege Research, Ine Reprinted with permission of Omega Research, Inc. PATTERN CODE UA830406-1 be 139T-BOND PATTERNS Day Session Only Pattern Code: UA830407-1 Type: Exact Position: Long Profit Target: 1 point = $1,000 Protective Stop: 1 point = $1,000 Order Entry: Open of Tomorrow, O Order delay: 0 Logic: C(0) > C(3) and C(1) > C(2) and C(3) > C(4) and C(2)>C(0) PATTERN CODE UA830407-1 140 T-BOND PATTERNS Day Session Only Pattern Code: UA830407-1 Historical Performance Results Period 2/15/83 - 1/7/99 TerFornance Samrary Highlights, Uiew-node! Options Total net profit Gross profit Total # of tredes Nonber winning trades Largest uinning trade fAuceage winning trade Ratio eug wiavaug loss fac closed-out dravdoun Profit factor Account size required ‘Systen Writer Plus $13,014.75 336,664 50 9-2,254.25 2.98 95,473.50 12/99 - at trades ear to toggle display Percent profits Nnber losing trades Largest losing trade fucrage losing trade fog trade (vin # loss) tex consecutive losers fhug © bars In losers. intracday dra ‘on account $-17.699.75 ” $1,236.75 $1,038.22 $358.77 z 2 9-2.473.50 1 Reprinted with permission of Omega Research, Inc. 141T-BOND PATTERNS T-BOND PATTERNS Day Session Only Day Session Only Pattern Code: UA830826-1 Pattern Code: UA830826-1 Type: Delay-Exact Position: Long Historical Performance Results Profit Target: 1 point Period 2/15/83 - 1/7/99 Protective Stop: 1 point Order Entry: Close, C(0) Order delay: 3 i Logic: H(4) > H(5) and H(3) > L(4) and L(5) > H(3) and L(4) > L(3) and H(5) > L(5) % View-nodel Options Space Total net profit $18,389.75, } Gross prose $35,820.75 Gross loss 32 431.00 53. Percent profitable om 36 Munber Iosing trades " $1,169.59 Largest losing trade $-1,143.00 $995.02 Average losing trade $1,025.35 0.97 fog trade (uin lose) 3346.58 fax consecutive winners 2 ax consecutive losers 2 fog bars in winners. fhug # bars’ In losers. 2 fax closed-out draudoun x intraday dravdoun —§-3,001.00 Provit factor ex fof contracts held fi Account size required . Return on account 306, Reprinted with permission of Omega Research, Inc. PATTERN CODE UA830826-1 142 143T-BOND PATTERNS T-BOND PATTERNS Day Session Only Day Session Only Pattern Code: UA830131-1 | Type: Exact Position: Short Pattern Code: UA830131-1 Profit Target: 1 point Protective Stop: 1 point Order Entry: Close, C(0) Order delay: 0 Period 2/15/83 - 1/7/99 Historical Performance Results Logic: C(3) > C(0) and C(1) > C(3) and C(2) > C(1) and c@) > CR) F Tonnes eam 7 Wighlights” Uiew-nodel Options Total vet profit $0,929.75 Gross profit $16,055.75 Gross toss Total # of trades a Nanber winning. trades eo Ranber 1 $1,325.75 Largest losing trade 31100345 fuerage losing trade fog trade Cul fog # bars in losers fax intea-day Nex # of cont Return on account Reprinted with permission of Omega Research, lnc. PATTERN CODE UA830131-1T-BOND PATTERNS T-BOND PATTERNS Day Session Only Day Session Only Pattern Code: UA830304-1 ' Type: Delay-Exact Position: Long Pattern Code: UA830304-1 Profit Target: 2 points = $2,000 Protective Stop: 2 points = $2,000 Order Entry: Close, C(0) Order delay: 1 Historical Performance Results Period 2/15/83 - 1/7/99 Logic: C(2) > C(4) and C(1) > C(3) and C(4) > C(1) TSTORTCAL HESULTS —Terfornance Summary Todet TF Equity Chart” Highlights” View-node! Options 12,99 ~ All trades Space bar to toggle display Totar net profit $25.340.25 Gross profit 343666 .50 Gross toss $19,318.25 Total # of trades 31 Tercent profitable Number uinning trades 22 Mander losing trades Largest ulmaing $2,044.50 Largest losing trade fAucrage winning 81,308.66 Ratio aug uinea 0.98 Nex consecutive winners 5 Max consecutive losers ‘ug bars in winners, 13 fog W°hars In losers Nox closed-out draudoun $4,170.50 fax intraday draudoun —$-4,170.50 Profit factor 2:30 fax # of contracts held 1 Account size required $7,170.50 Return on account, a8 Reprinted with permission of Omega Research, Inc. PATTERN CODE UA830304-1 146 147T-BOND PATTERNS Day Session Only Pattern Code: UA830125-1 Type: Exact Position: Long Profit Target: 0.5 points = $500 Protective Stop: 0.5 points = $500 Order Entry: Close, C(0) Order delay: 0 Logic: H(1) > L(0) and H(0) > L(2) and H(2) > H(Q) and L() > L(1) and L(2) > H(1) PATTERN CODE UA830125-1 148 T-BOND PATTERNS Day Session Only Pattern Code: UA830125-1 Historical Performance Results Period 2/15/83 - 1/7/99 ferTarnance Sranary Equity Chart Highlights Uiew-nodel Options unnary, Detailed, Listing 1299 - Ah Tet ® 1of Space Total net profit E | Grass profit $12:658 00 Gross lose Total # of trades 2% Percent profitable Nunber-vinning trades 19 Minber Losing trades Largest winning trade $2,369.50 Largest losing trade i foctage wioning $666.21 moctngs losing trade | Rettolsuy sinter Tose 12 fog trean tain A'iossy 5320.38 hax consecutive winners © fax consecutive losers 2 fog # bers in ul 1 fg W hers In losers: 1 Profit. factor fax closed-out éraudoun Nex incra-day draudoun— §-1,090.00 Account size required 20% Systen Uriter Plus ——— Oneye Research, Ine. Reprinted with permission of Omega Research, Inc. 149T-BOND PATTERNS Day Session Only Pattern Code: UA830418-1 Type: Exact Position: Short Profit Target: 0.5 point Protective Stop: 0.5 point Order Entry: Close, C(0) Order delay: 0 = $500 Logic: H(3) > L(2) and H(2) > L(0) and L(1) > H(3) and H(0) > H(1) and L(2) > L(3) and L(0) > L(1) and H(1) > H() PATTERN CODE UA830418-1 150 T-BOND PATTERNS Day Session Only Pattern Code: UA830418-1 Historical Performance Results Period 2/15/83 - 1/7/99 Terforsance Saneary ‘Chart Highlights View-nodel ‘Space bar to toygle display Total net profit $6,052.50 Gress protlt, 911,255.50 Gross loss of trades 25 inning trades a Largest winning trade $638.25 Aucrage uinning trade $409.61 poe Ratio aug wineaug Loss 0.95, tox consecative winners 4 ‘hug # bars in w i Mex closed-out dravdown $-1,072.60 Protit. factor L system uriter Pius ‘Dnege Research, Ine Reprinted with permission of Omega Kesearch, Inc. 151T-BOND PATTERNS Including Evening Session Pattern Code: US800310-1 Type: Exact Position: Long Profit Target: 1 point = $1,000 Protective Stop: 1 point = $1,000 Order Entry: Close, C(0) Order delay: 0 Logic: C(0) > O(2) and O(0) > C(1) and C(2) > O(1) and (1) > C(2) and O(2) > O(0) PATTERN CODE US800310-1 152 T-BOND PATTERNS Including Evening Session Pattern Code: US800310-1 Historical Performance Results Period 1/1/80 - 1/7/99 Gross profit Total # of trades under vinving trades Largest winning trade fucrage winning trede Ratio aug winvaug’ loss flex consecutive wingers fog # bors in winners Provit tector Account size reqeired Syston uriter Plus 31,013.25, $503.43 098 5 4 2,206.50 2.51 $5,447.09 Largest losing trade foctage losing trade fog trade (uit & toss) tar consecutive losers ug # bars in losers Max t"of contracts held Return on account, Drege Hesearch, Ine. 90,297.75 mm é g-teait.75 $4029 72 S4za 12 2 2 92.447 00 2 Reprinted with permission of Omega Research, Inc. 153T-BOND PATTERNS Including Evening Session Pattern Code: US801701-1 Type: Exact Position: Short Profit Target: 1 point = $1,000 Protective Stop: 1 point = $1,000 Order Entry: Close, C(0) Order delay: 0 Logic: C(3) > C(O) and C(2) > C(1) and C(4) > C(2) and C(I) > C3) PATTERN CODE US801701-1 154 T-BOND PATTERNS Including Evening Session Pattern Code: US801701-1 Historical Performance Results Period 1/1/80 - 1/7/99 TSTORTCAL RESULTS FerFornance Sunnary Equity Chart Highlights” lew-nodel Options led, Listing 1299 ~ al trades: ace ber to toggle display Gross loss $-3,349.50 Percent prof Nunber winning trades omber losing Largest winning trade 02 Largest Losing tra focrage winning trade fverage losing Ratio aog fog trade Coin fax consecutive winners 10 tax consecutive losers fog bars in winners, ‘3 hug # bars in losers fax closed-out draudoun $3,126.00 Max Intre-day draudoun —§-3, 160.50 Profit factor, 1.89 Max # Of contracts held 1 Account size required 96,100°59 Return on account 13 Systen ueiter Plus ‘Onega Research, Inc Reprinted with permission of Omega Research, Inc. 155T-BOND PATTERNS T-BOND PATTERNS Including Evening Session Including Evening Session Pattern Code: — US810220-1 Type: Exact Position: Short Profit Target: 1 point = $1,000 Protective Stop: 1 point = $1,000 Order Entry: Open of Tomorrow, O Order delay: 0 Pattern Code: US810220-1 Historical Performance Results Period 1/1/80 - 1/7/99 Logic: C(1) > H(2) and C(2) > L(1) and H(0) > C(0) and i FerTarmance Sammon fed, Listing H(@) > L(0) and H(1) > C(1) and L(0) > C(2) and eR chart nant abte etel Options C(0) > H(1) and L(1) > L(2) FP eer eee ee eee 12799 ~ al tras Space Total set profit, Gross provi Grose lose Percent profitable Masher vival IMinber losing trades anber winning 9902.00 losing trade fucrage winning trade Sonz.00 osing trade : 4 2.00 Largest winning tr Ratio aug winvayg Loss (ain a loss) reutive losers sin losers. held it! PATTERN CODE US810220-1 Reprinted with permission of Omega Research, Inc. 156 157T-BOND PATTERNS T-BOND PATTERNS Including Evening Session i Including Evening Session Pattern Code: US800111-1 Type: Exact Position: Short Profit Target: 2 points = $2,000 Protective Stop: 2 points = $2,000 Order Entry: Close, C(0) Order delay: der delay 0 Period 1/1/80 - 1/7/99 Pattern Code: US800111-1 Historical Performance Results Logic: C(2) > C(0) and C(3) > C(2) and C(1) > C(3) options 29 All trades Space Sar to togaie dlepley Total wet profit 342,664.25, Gross pratt 390,602.50 Gross loss 947,330.25 Total # of trades 69 Percent profitable under winning trades 45 Mnber losing trades Largest winning trade 92.544.50 fecrege winning trade 321000106 fuer. og winveug. loss tor ° Nex closed-out draudoun — $-4.324.00 Profit factor 190 fax # of Acenunt site required $2,942.25 Return on Reprinted with permission of Omega Research, Inc. PATTERN CODE US8001 11-1 158 159CRUDE OIL PATTERNS 161CRUDE OIL PATTERNS CRUDE OIL PATTERNS Pattern Code: CL840221-1 Type Exact Position: Long Pattern Code: CL840221-1 Profit Target: 1 point = $1,000 Protective Stop: 1 point = $1,000 Order Entry: Close, C(0) Historical Performance Results Order delay: 0 Period 2/15/84 - 1/7/99 Logic: L(1) > H(2) and H(1) > L(0) and H(3) > L(2) and L(2) > L(3) and L(0) > L(1) and H(0) > H(1) and TOAL RESULTS ~ Performance 7, H(3) > LQ) af seity Chere Rigmighte ewvoiel options 1299 — Alle Spat {$13.650.00 324/958.09 Gross toss 35 Percent profitable winning trades 24 Number Tosing trades J uergest winning trade $1,512.00 Largest losing trade focrage winning trade $1,039.92 Aucrage Losing trade Ration aug uincaug. loss 102 fug trade Coin & Loss) tex consecutive winaers 6 Max consecutive losers fog # bars. In win 14 hug # bare in losers, _ Lax of contracts held Account size required Return on account, Reprinted with permission of Omega Research, Inc. PATTERN CODE CL840221-1 162 163CRUDE OIL PATTERNS CRUDE OIL PATTERNS Pattern Code: CL850210-1 Type Exact Position: Long Profit Target: 1 point = $1,000 Protective Stop: 1 point = $1,000 Order Entry: Close, C(0) Order delay: 0 Pattern Code: CL850210-1 Historical Performance Results Period 2/15/84 - 1/7/99 Logic: C(2) > C(1) and C(0) > C(3) and C(1) > C(4) and (3) > C2) H Total vet profit $9,528.00, Gross profit $17,224.00 Gross lose Total # of trades 24 Percent profitable Humber winning trades 4? Humber losing trades Largest winning trade $tstez.eo Largest losing trade inning trade $1,013.19 uerage losing trade Matiotaug vincevy Toss (0.92 Aug trade Cain & Loss) ex consecutive winners 6 tax consecutive losers ‘hug # bars in winners 11 fg # bers in losers cloced-out drawdown 9-2.092.80 fax intraday Araudoun Frorit factor 2.24 fax fof contracts held ]] hccount size required $ Return on account Reprinted with permission of Omega Research, Inc. PATTERN CODE CL850210-1CRUDE OIL PATTERNS CRUDE OIL PATTERNS Pattern Code: CL850919-1 Type Exact | Position: Long Profit Target: 1 point = $1,000 Pattern Code: CL850919-1 Protective Stop: 1 point = $1,000 Order Entry: Close, C(0) Order delay: 0 Historical Performance Results Period 2/15/84 - 1/7/99 Logic: H(1) > H(2) and L(0) > L(1) and H(0) > C(O) and H(2) > O(0) and O(0) > L(0) and L(0) > L(1) and LQ) > L(2) TFerTopanee Toa ry n Chart Highlights” View-nogel Options Rusnary, Detailed, Listing 12/99 - Al trades ‘Space bar to toggle display $13,416.00 326,362.00 Gross loss $12,946.00 Total # of trades 38 Fercent profitable Nanber winning trades 26 Manber losing trades Largest winning trade $1,332.00 losing trade fverege winning trede $1,013.92 Ratio aug ulnzeug loss tex consecutive winners fine consecutive tos 2 fy # bars. In winners fog # bars In loser a Niax closed-out draudoun dey dravdoun — §-2,668.00 Protit factor, lontracts hel Account tize required Return on account 2362 Reprinted with permission of Omega Research, Inc. PATTERN CODE CL850919-1CRUDE OIL PATTERNS Pattern Code: CL841206-1 Type Exact Position: Long Profit Target: 1 point = $1,000 Protective Stop: 1 point = $1,000 Order Entry: Open of Tomorrow, O Order delay: 0 Logic: H(2) > C(2) and L(0) > H(2) and H(0) > H(1) and C(O) > L() and H(1) > C(1) and C(2) > L(1) and LQ) > L() and C(1) > C(O) PATTERN CODE CL841206-1 168 CRUDE OIL PATTERNS Pattern Code: CL841206-1 Historical Performance Results Period 2/15/84 - 1/7/99 a Biew-nadel Options a 12/99 ~ All trades —————— 1 Space bar to toggle @uiay Total net profit Gross profit Gross loss Str 90 Total # of trades 32 Percent profitable Nunber vineing trades 21 Munber losing tredes 1 Largest winning trade $1,182.00 Largest tosing trade fuerage uinning trade $4,009.62 Average losing trade Ratio aug winzavg toss 9 ug trade (uin & loss) tex consecutive vinners 7 Max consecutive losers 1 fog # bars In winners 12 fg # bars in losers “ fax closed-out ara Profit factor Account size required foun $-3,030.00 Max lay drewdoun 9 4.110 00 93 Max fof contracts held $6. 10_Return on account tu ‘nega Research, Ine. — Copyright tit — Systen writer Plus Reprinted with permission of Omega Research, IncSWISS FRANC PATTERNS 171SWISS FRANC PATTERNS Pattern Code: — SF830520-1 Type Exact Position: Short Profit Target: 1 point = $1,250 Protective Stop: 1 point = $1,250 Order Entry: Close, C(0) Order delay: 0 Logic: C(1) > C(0) and C(2) > O(1) and O(0) > C(1) and O(2) > C(2) and O(1) > O(0) PATTERN CODE SF830520-1 172 SWISS FRANC PATTERNS Pattern Code: SF830520-1 Historical Performance Results Period 2/15/83 - 1/7/99 [Performance Sunnary MIG equity chart” Highlights Uiew-nodel options ry, Detatled, Listing 1299 ~ All trades Space ber to toggle display Total net profit 913,736.00 Gross profit §28122050 Gross toss Total # of trades 28 fercent profitable nk snming trade 13 Number Tosing tredes $2,057.00 Largest losing trade $4/327139 fuerage losing trade Ratio aug ui fog trade (win & loss) hex consecutive winners Nex consecutive losers fog 8 bare. In fog # bars in losers Reprinted with permission of Omega Research, Inc. 173SWISS FRANC PATTERNS SWISS FRANC PATTERNS Pattern Code: — SF831011-3 Type Matched oo a eel Pattern Code: SF831011-3 Profit Target: 1 point = $1,250 Protective Stop: 1 point = $1,250 Historical Performance Results Order Entry: Open of Tomorrow, O Period 2/15/83 - 1/7/99 Order delay: 0 Logic: C(3) > C(2) and C(2) > C(1) and C(4) > C(0) and Tianary (CQ) > C(3)) or (CO) > C(2) and C(3) > C(4))) Pete Peraeaaeeeaeaeto ere 3299 - all Space bar to toggle display EXACT PATTERN | EXACT PATTERN 2 Total vet profit $17.467.50 Gross profit, 335,051.50 Gross loss 917,584.00 6% Total # of trates 40 Fercent profitable ates a funder winning trades 27 fanber 105) winning trade $1,282.00 Largest losing trade §-1,730.50 je winning trade $1,296.20 Average Tosing trade tog uinvaug. loss 10.96 fug trade (uin & loss) cutive vinners 5 tax consecutive losers rin winners Sug # bars in losers $1730.58 $5,305.50 Reprinted with permission of Omega Research, Inc. PATTERN CODE: SF831011-3 174 175SWiSS FRANC PATTERNS Pattern Code: SF830519-1 Type Delay-Exact Position: Long Profit Target: 1 poin Protective Stop: 1 point Order Entry: Close, C(0) Order delay: 3 Logic: C(7) > C(3) and C(6) > C(5) and C(4) > C(7) and (5) > C(4) PATTERN CODE: SF830519-1 176 SWISS FRANC PATTERNS Pattern Code: SF830519-1 Historical Performance Results Period 2/15/83 - 1/7/99 TerTarnance Summary hart. Highlights” Ulewnode! Options 12/99 ~ @) trades Space bar to toggle display J} toeai vet prosie {20,011.50 Grose profit $0400.50 Gross loss Total # of t 32 Percent profitable Hunter winni umber losing trades Largest ulmning ij trade Average vinning fuerage losing trade Ratio. aug uinvaug fog trade Cain & Toss) tax consecutive winners hax consecutive losers fog # bars In losers Reprinted with permission of Omega Research, Inc 177S&P 500 PATTERNS 179Pattern Code: Type Position: Profit Target: Protective Stop: Order Entry: Order delay: S&P 500 PATTERNS SP830825-1 Exact Long 10 points 10 point: Open of Tomorrow, O 0 Logic: H(Q) > L(1) and L(2) > H(0) and H(1) > L(2) and H(2) > L@) and L(1) > L(0) and L(3) > H(1) and H(3) > H(2) PATTERN CODE SP830825-1 180 Total net prorit Gross profit Total # of trades Minber winning trades S&P 500 PATTERNS Pattern Code: SP830825-1 Historical Performance Results Period 2/15/83 - 1/7/99 FerFormance Sunnary Highlights, Uiew-nodet Options 12/99 - atl trades Space bar te toggle display 922,576.50 $47,720.50 Gross love $20,144.00 22 Percent profitable 20% {5 Number losing trades 8 Largest winning trade $2,982.00 Largest losing trade hocrage winning trade 82,511.61 fuerege losing trade Ratio aug winsaug loss 160 fg je'tuin 8 loss) tex consecutive winners hug # bars in winners, flax closed-out draudoun Provit factor, Account size required 6 we $-7,554.00 2197 $24,941:50 hex consecutive losers fhug © bars In losers Araudoun — $-7,941.50 acts held 1 Reprinted with permission of Omega Research, Inc. 181S&P 500 PATTERNS Pattern Code: — SP830120-1 Type Exact Position: Long Profit Target: 10 points = $2,500 Protective Stop: 10 points = $2,500 Order Entry: Open of Tomorrow, O Order delay: 0 Logic: H() > C(0) and L(2) > L(0) and H(1) > H@) and H(2) > H(1) and L(0) > L(1) and C(0) > L(2) PATTERN CODE SP830120-1 182 S&P 500 PATTERNS Pattern Code: SP830120-1 Historical Performance Results Period 2/15/83 - 1/7/99 TSTORTCAL WESULTS ~ Performance Sunnary Equity Chart Highlights Uiew-nodel Options ye Detatted, Listing Space ber to toggle 4isplay Total net profit $34,307.50 Gross profit 399171150 Gross loss $-45.324.00 so 2 fax consecutive winners fog # bers in winners: Nex closed-out draudoun Protit factor Account size requ Reprinted with permission of Omega Research, Inc. 183S&P 500 PATTERNS Pattern Code: SP830330-1 Type Exact Position: Long Profit Target: 10 points Protective Stop: 10 points = $2,500 Order Entry: Open of Tomorrow, O Order delay: 0 Logic: H(0) > C(0) and L(2) > L(1) and H(1) > H(2) and H(2) > L(0) and C(0) > H(1) and L(0) > L(2) PATTERN CODE SP830330-1 184 S&P 500 PATTERNS Pattern Code: SP830330-1 Historical Performance Results Total # of trades under winning trades Largest stoning trade fsverege win Ratio avg winvaug loss tax consecutive vinsers fog # bars in winners Nox closed-out draudoun Procitfactar Account size required Period 2/15/83 - 1/7/99 Gross toss 32 Fercont protien 26 © Munber Iosing trades $2-719-50 Largest losing trate $21452 58 tuerage 197 hug trade (eth & loss) tex consecutive losers 11 fog # bars in losers $-5.036.00 ex intra~day draudoun — $-5,190.50 ex ¥ of contracts held $22,190.50 Return on account Reprinted with permission of Omega Research, Inc. 185SILVER PATTERNS 187SILVER PATTERNS Pattern Code: — S1830520-1 Type Exact Position: ‘Short Profit Target: 40 points = $2,000 Protective Stop: 40 points = $2,000 Order Entry: Open of Tomorrow, O Order delay: 0 Logic: H(0) > C(1) and L(2) > H(0) and H(1) > C(2) and CQ) > L(2) and L(1) > C(O) and C(O) > L(0) and H(2) > H(1) and C(1) > LQ) PATTERN CODE. $1830520-1 188 SILVER PATTERNS Pattern Code: $1830520-1 Historical Performance Results Period 2/15/83 - 1/7/99 ToT RESULTS PerTorsance Sanmary Tyuity Chart Highlights Uiew-nodel Options Riemer, Detailed, Listing 1299 - Short ‘Space ber to toggle display Total net profit Brose profit Gross toss Total # of trades Percent profitable Hunber winning trades Nomber 1osing trades $2,497.00 Lergest losing trade 3202200 fuerage losing trade Ratio aug wlnsavg. loss 100 fug trade (win & Loss) tex consecutive winners 5 ex consecutive losers fog # bars in winners, 44 hug # bars in losers tex closed-out dravdown §-4,072.00 Mex Intea-day draudoun $-5.061.00 Provit factor 2149 Mex # of contracts held 1 Account size required $8,061.00 Return on account Reprinted with permission of Omega Research, Inc. 189Pattern Code: Type Position: Profit Target: Protective Stop: Order Entry: Order delay: SILVER PATTERNS $1830630-1 Delay-Exact Short 20 points = $1,000 20 points = $1,000 Open of Tomorrow, O 3 Logic: C(4) > C(3) and C(6) > C(4) and C(7) > C(6) and CQ) > CB) PATTERN CODE: $1830630-1 190 SILVER PATTERNS Pattern Code: $1830630-1 Historical Performance Results Period 2/15/83 - 1/7/99 TAL TESTLTS + PerTornance Sunnary r ity Chart Mighlight®, ary, Detailed, Listing View-nodel options 12/99 - All trades Space bar to toggle display Total net profit, 99,930.00, Gross profit 922177100 Gross loss $12,841.08 Total # of trades flanber winning. tr Largest winning trade fucrage 4 Ratio avg winseug loss tex consecutive winners fog @ bars [1 winners. hax elosed-ou* ar Protit factor Account size required Systen writer Plus Lergest losing trade fverage, los fg trade Cain tex consecutive lorers fog # bars In losers Nex intra-dey drawdown §-3,097.00, ex # of contracts held 1 Return on account 162 Reprinted with permission of Omega Research, Inc. 191INTER-MARKET PATTERNS 193S&P 500 PATTERNS LONG T-BONDS, DAY SESSION S&P 500 PATTERNS LONG T-BONDS, DAY SESSION Pattern Code: SP830128-2 Type Exact Position: Long T-BONDS Profit Target: 1 point = $1,000 Protective Stop: 1 point = $1,000 Order Entry: Open of Tomorrow, O Order delay: 0 Pattern Code: SP830128-2 Historical Performance Results Period 2/15/83 - 1/7/99 Logic: H(2) > C(2) and L(0) > L(1) and H(0) > H(1) and C(0) > L(1) and H(1) > C(1) and C(2) > L(2) and L() > H(2) and C(1) > C(0) a Séew-nodel Options 1299 - @l trades Space bar to toggle display Total vet profit $11,085.25 Gross prorit 316,975.25 Gross loss Total # of trades 22 Percent inter vineing trades v Largest winning trade $1,232.00 Largest losing trade fuerage $990.54 fucrage losing trade $1,018.00 Ratio aug vinvaug. toss ‘0.88 hug trade (win & loss) $540.24 PATTERN CODE SP830128-2 194 ex consecutive winners fog # bers In winners tex closed-out drawdown Profit. factor, Account size required hax consecutive losers fog # bars" in losers tex intraday ‘4 fax fof contr Reprinted with permission of Omega Research, Inc. 195S&P 500 PATTERNS LONG T-BONDS, DAY SESSION Pattern Code: — SP830107-2 Type Exact Position: Long T-BONDS Profit Target: int = $1,000 Protective Stop: Order Entry: Open of Tomorrow, O Order delay: 0 Logie: H(2) > C(2) and L(0) > H(2) and H(0) > H(1) and C(0) > C(1) and H(1) > C(0) and C(2) > L(1) and L(1) > L(2) and C(1) > L(0) PATTERN CODE SP830107-2 196 S&P 500 PATTERNS LONG T-BONDS, DAY SESSION Pattern Code: SP830107-2 Historical Performance Results Period 2/15/83 - 1/7/99 TSTORTCAT_WESULTS 7 Perfornance Summary ‘Chart. Highlights View-roéel options ry. Detaled, Listing rade = Space bar to toggle aizplay Total net profit 98,768.25, Gross provit 917163150 $-0,962.25, of trades 25 Percent profitable 6B inning. trades ¥ —famber losing trades ® $1,825.75 Largest losing trade 31105215 fuerage losing trade 4 fog trade (win & Loss) tex consecutive winners & Max consecutive losers fog # bars in winners 2 fax closed-out drawdown $2,629.75 Profit factor Account size required 95.73.50 Reprinted with permission of Omega Research, Inc. 197S&P 500 PATTERNS S&P 500 PATTERNS LONG T-BONDS LONG T-BONDS Pattern Code: SP830825-2 Type Exact Position: Long T-BONDS Pattern Code: SP830825-2 Profit Target: 2 points = $2,000 Protective Stop: 2 points = $2,000 Historical Performance Results Order Entry: Open of Tomorrow, O Period 2/15/83 - 1/7/99 Order delay: 0 Logic: H(2) > L(3) and L(2) > H(0) and H(1) > L(2) and DEED "Equity. chart” Wightights Otew-nodet Options H(0) > L(1) and L(3) > H(1) and H(@3) > H(2) and ener Sevag Viste LQ) > L(0) ws 99) i sce bar to toggle alsplay Total net profit $17,942.00 Gross profit. $40,171.25 Total # of trades Number winning trades 2 2 ing trade $2,263.25 Largest losing trade ing trade $2,008.56 fuerage losing trade ig wineug loss 0.93 fog trade (win & loss) tex consecutive winners 2 fax consecutive losers ug # bars in winners. 14 fog B bars, Hy $4,036.00 Max intravday drawdown fit. factor fl fax @ of contracts held Recount ‘size’ require $9,317.25 Return'on account Reprinted with permission of Omega Research, Inc. PATTERN CODE SP830825-2 198 199S&P 500 PATTERNS LONG T-BONDS, DAY SESSION Pattern Code: $P830405-2 Type Exact Position: Long T-BONDS Profit Target: i Protective Stop: Order Entry: Open of Tomorrow, O Order delay: 0 Logic: H(1) > L(1) and LQ) > L(0) and H(0) > H(1) and H(2) > H(0) and L(1) > L@) PATTERN CODE SP830405-2 200 S&P 500 PATTERNS LONG T-BONDS, DAY SESSION Pattern Code: SP830405-2 Historical Performance Results Period 2/15/83 - 1/7/99 Equity chart nary» Detailed, List ing $14,602.75 $320.747.00 Gross loss 1 8 of trades 27 Fercent profit Nonber winning trades 24 Minber losing 91,107.00 $987:95 nverage losing trade ° Ratio avg winvauy. loss 3? fug trade (win & loss) fax consecutive winners 9 fax consecutive losers fog # bars In winners. fiug W bars in losers, Reprinted with permission of Omega Research, Inc. 201S&P 500 PATTERNS S&P 500 PATTERNS LONG S&P 500 LONG S&P 500 Pattern Code: US881011-2 Type Exact : : Position: Long S&P 500 Sati eats Profit Target: 10 points = $2,500 Protective Stop: 10 points = $2,500 Historical Performance Results Order Entry: Open of Tomorrow, O Period 2/15/83 - 1/7/99 Order delay: 0 Logic: C(0) > C(3) and C(2) > C(1) and C(1) > C(O) TEAL WERT > YerToranss Ty cast ayclen Equity hort ys Detalled, Listing 1299 ~ AL trades Space bar to toggle disp Total net profit 950,965.00 Gross prorit $118,223.00 Gross loss $-59.264.00 Total # of trade: 2 Percent profitable or 2 Mander winning tredes 4? fanber Losing trades Largest winning trade $3,957.00 Largest losing trade $3,043.00, fvcrege winning trade $2,515.51 average losing trade Ratio avg vinraug lors 0.56 fug trade (uin & loss) tex consecutive winners 9 tax consecutive losers fog # bers in wlnners. 13, fg # bers In losers $-7,626.00 Mex closed-out draudoun —$-7,626.00 Mex intraday draudoun Protit factor 99 fax of contracts held Account size required $17,626.60 Return on account Reprinted with permission of Omega Research, Inc. PATTERN CODE US881011-2 202 203SWISS FRANC PATTERNS SHORT SILVER Pattern Code: SF850419-2 Type Exact Position: Short COMEX Silver Profit Target: 20 points = $1,000 Protective Stop: 20 points = $1,000 Order Entry: Open of Tomorrow, O Order delay: 0 Logic: L(1) > L(2) and H(1) > H(0) and H(2) > L(1) and H(0) > L(0) and L(0) > H(2) PATTERN CODE SF850419-2 204 SWISS FRANC PATTERNS Hist SHORT SILVER Pattern Code: SF850419-2 orical Performance Results Period 2/15/83 - 1/7/99 Perfor ‘chart Listing Total vet proie Gross profit 4 Ratio aug ulnvaug loss fax consecutive winners fog bars: In winners hex closed-out drawdown Provit factor Account size required = Sanmary Highlights, Ulew-nodel Options 1299 - #1 trades ‘Space bar to toggle display $13.963..00 824114300 Gross loss 3-10,190.00 4 20% a $1,282.00 $1,005.96 0199 tug trade (ulm & toss? $410.65, ° tive losers 2 10 Tosers 6 $-2,119.00 raudown — §-2,611.00 299 t 95,611,00 2a ‘Onega Research, Ine Reprinted with permission of Omega Research, Inc 205INDEX 1 A Abstract models, 24 indicator predicting power, 119 Cucra oF Indicators,7 smoothing, 119 7 interaction ratio, 59 back testing, 34 intraday trading, 5 c L Bors Long-term trading, 10 Charting methods, 14 consecutive losers, 36 M continuously adjusted data, 18 margin safety factor, 106 requirement, 105, D ‘mathematical model, 25, daily report, 90 model, 24 cds daily operation, 60 7 implementation, 33, simulation, 34 Easy Language, 41; 44; 53; 65; 96 validation, 35, empirical models, 24 verification, 34 modeling, 24 Money Management, 101 entry points, 26 N F next day projection technique, 62; 4% fitting, 33, NYMEX Crude Oil futures, 90 fundamental trading methods, 13 ° 6 ‘Omega Research, In, 42 Good “Til Canceled, 63 pene ; output states, 25 4 P Hindsight, 10 pattern Historical Price Patterns, 9; 15, Rasen ie definition, 73, delay, 77 exact, 73 207identification algorithm, 86 inter-market, 79 matched, 75 proportional, 75 search, 83 split,77, split-matched, 79 prIndicator, 54; 120; 124; calculation, 123 characteristics, 123, construction, 121 definition, 120 Q quarterly returns, 43, R risk, 101 constant, 102 full, 117 reduced, 117 variable, 102 Robbins Trading Company, 54 s short term trading, 8 simulation, 28 Slippage.6 statistics, 29 stochastic process, 25, symbolic mathematics manipulation, 99 T T-Bond Futures, 39 technical trading methods, 14 temporal aggregation, 35, trading signals clustered, 112 208 coincident, 110 successive, 109 ‘Trading Time Frame, 5 trend following, 113 v visual search, 56 w World Cup Championship of Futures Trading, 54 Reader Service Information More information can be obtained as follows: Web Site: www.tradingpatterns.com By Fax: (212) 898-9076 By e-mail: mharris@tradingpatterns.com By mail: Address letters to: Mike Harris 1204 Third Avenue Suite 251 New York, NY 10021 Please include your name and telephone or Lay tnuimiber 209WW.TRADINGPATTERNS.COM TRADING PATTERNS (“ome — EE A UNIQUE SERVICE FOR STOCK AND FUTURES TRADERS ul ie BUY {ERE TO LEARN MORE! IF YOU ARE A DAY TRADER OR A SHORT TERM ‘TRADER, FIND OUT HOW YOU CAN USE PRICE PATTERNS TO PROFIT FROM THE STOCK AND. FUTURES MARKET! Please contact our webmaster with questions or comments. © Copyright 1999 Haern Holdings, La Alleghts reserved