Designing Stock Market Trading Systems

With and without soft computing

Bruce Vanstone, Tobias Hahn

• Verlag: Harriman House
• Jahr: 2011
• ISBN:
  9780857191359
• Auflage: 1
• Thema: Business
• Thema: Stocks

In Designing Stock Market Trading Systems Bruce Vanstone and Tobias Hahn guide you through their tried and tested methodology for building rule-based stock market trading systems using both fundamental and technical data. This book shows the steps required to design and test a trading system until a trading edge is found, how to use artificial neural networks and soft computing to discover an edge and exploit it fully.Learn how to build trading systems with greater insight and dependability than ever beforeMost trading systems today fail to incorporate data from existing research into their operation. This is where Vanstone and Hahn's methodology is unique. Designed to integrate the best of past research on the workings of financial markets into the building of new trading systems, this synthesis helps produce stock market trading systems with unrivalled depth and accuracy.This book therefore includes a detailed review of key academic research, showing how to test existing research, how to take advantage of it by developing it into a rule-based trading system, and how to improve it with artificial intelligence techniques.The ideas and methods described in this book have been tried and tested in the heat of the market. They have been used by hedge funds to build their trading systems. Now you can use them too.

Chapter 1 –Designing Stock Market Trading Systems

“Tips are for waiters.”

Daryl Guppy

1.1 Introduction

This book is designed to assist the reader with two main fields of study, namely:

1. Designing mechanical stock market trading systems, and
2. Using soft computing to enhance mechanical stock market trading systems.

This book presents a defined methodology for creating financially viable mechanical stock market trading systems, both with and without soft computing.

The design methodology presented here sees the building of mechanical stock market trading systems as the assembly of a number of tested components. This book explains the function of these components, and relates them to previously published academic research. Further, it demonstrates appropriate metrics to test individual components, and also to test the finished product; the trading system.

The process of building mechanical stock market trading systems is not a linear one. Although each function can be studied separately, the resultant trading system is a complex interaction between all the components, with the operation of each component affecting the others (and perhaps even itself).

For this reason, it is necessary to study each of the main functions separately, and have a solid understanding of what each function is designed to do. When this objective has been reached, it is then time to start considering the interactions between the functions. When all the functions are working together in harmony, the result is a highly usable trading system.

1.2 Motivation

A primary motivation for writing this book was to provide aspiring trading system developers with a design methodology to follow. Many approach the task as if it were a ‘black art’, endlessly thrashing with a vast array of technical indicators in the unlikely hope of suddenly pulling a successful system out of the chaos. There is no justification for this style of system development, which eventually becomes bogged down in data mining and curve-fitting until it finally ends with analysis paralysis. ^[1]

Designing successful stock market trading systems is a complex, nonlinear process. The methodology presented in this book was the one I created and defined in my own PhD journey, which eventually culminated in my PhD thesis, ‘Trading in the Australian Stock Market using Artificial Neural Networks’. ^[2]

A defined methodology breaks a complex process up into smaller, more manageable pieces, and then allows a way for each piece to be specified and tested against those specifications. Finally, it provides a way to assemble the pieces, and to test the finished product.

Without methodology, a designer is lost. The design process becomes akin to an infinite number of monkeys at the keyboard. There is no real progress except by luck. There is no way to evaluate intermediate results, as there is no real understanding of their contribution to the overall goal.

The first step is to break this cycle, and begin the process of design. To do this, it is essential to understand the goal, the components that need to be built, and the constraints that exist.

1.3 Scope and data

The methodology presented in this book is applicable to many markets, instruments and traders.

Dependant on the example, it may use fundamental and/or technical data, which was sourced under the following conditions:

Technical data (open, high, low, close, volume) was sourced from Norgate Financial Services (2008). The technical data in the case studies in Chapter 10 covers ordinary shares in the S&P ASX200 (2004), during the time frame 3 April 2000 (index creation) to 31 December 2008. The data includes delisted stocks and is adjusted for dilutions, stock splits, and consolidations.

Fundamental data was sourced from the Aspect Financial Services database FinAnalysis (2004). This source provides detailed fundamental information for all companies listed on the ASX, and also maintains this data for delisted companies. In this book, the use of all fundamental data has been delayed for six months from the end of the financial year, which is (due to reporting requirements), well after the publication date. We store our fundamental data in a Microsoft Access database, and you will see an include named ‘Book_QFDBLib2’ in our examples that use fundamental data. This include is our way of connecting to the fundamental data we have stored in Microsoft Access. To repeat our examples when you are investing, you will need to replace this and the relevant fundamental data accesses in our examples with whatever code is required for you to connect to your own fundamental data source.

Where simulations are given, transaction costs are always taken into account. Transaction costs of $50 each way are used, and simulations also include slippage of 0.5%.

It is assumed that you will want to repeat the case studies in your own investing and, indeed, you are encouraged to do so. You should be aware that the S&P ASX200 was chosen for these examples mainly due to its high liquidity. If you are repeating these examples with different stock indices you should ensure there is adequate market depth to enable you to actually place trades (particularly exits). This caution is especially important if you are trying to trade amongst the penny stocks, which are notorious for being illiquid.

1.4 The efficient market hypothesis

The issue of market efficiency, or what is known in the popular press as the random walk theory, is one of the most hotly debated and thoroughly examined areas within the field of finance. Yet, market efficiency still remains a hypothesis, and volumes of well respected publications gather increasing amounts of evidence against it.

An interesting early review of some of this evidence is provided by Lehmann, who claimed it is now “open season on the efficient market hypothesis” and a comprehensive review of deficiencies in the efficient market hypotheses was then provided by Haugen. ^[3] There is also speculation amongst some academics as to whether efficiency is in some way related to the maturity of the market itself. Los added some credibility to this suggestion, by performing nonparametric testing on all six major Asian stock markets and finding them all to be inefficient, and that none exhibited random walk behaviour. ^[4]

Since the early work of Fama, it is common to discuss three forms of efficiency when looking at the efficient market hypothesis (EMH). ^[5]

1. A market is weak form efficient if it is not possible to consistently earn excess returns using past prices and returns.
2. A market is semi-strong efficient if it is not possible to consistently earn excess returns using any public information.
3. A market is strongly efficient if it is not possible to consistently earn excess returns using any information, including private information.

From a trading point of view, market efficiency in general means that it is not possible to consistently earn excess returns using any available information. In essence, then, the only thing that causes security prices to change is new information. By its definition, the arrival and timing of new information is unpredictable. Therefore, in an efficient market, security prices should appear to be generated randomly, hence the term random walk. It’s the academic equivalent of saying you never know what’s coming around the next corner.

There is also an interesting and obvious paradox here, in which the market could only ever be efficient if some investors believed it to be inefficient. ^[6] In other words, if all participants believed the market to be efficient, there would be no incentive to seek out new information and, therefore, no ability for this information to be assimilated into share prices.

It is not the intention of this book to chronicle the progress of the EMH debate. It is the intention of this book to arm you with the tools and knowledge you need to design stock market trading systems, both with and without soft computing. After that, you will undoubtedly make your own decision regarding the plausibility of the EMH.

1.5 The illusion of knowledge

Many investors spend an inordinate amount of time trying to find more and more information. They seem to inherently believe in the EMH, by acting as though they believe the only way to outperform others is to acquire more information than them. However, there is a great deal of published research that demonstrates that the accuracy of human decision-making does not increase in proportion with the amount of information available. ^[7]

An experiment that provided evidence for this involved MBA students from an advanced finance course who were asked to forecast stock earnings. ^[8] Three sets of data were used: baseline data, baseline data plus redundant news information, and baseline data plus non-redundant news information....