Part IV

Building advanced models without proper data preprocessing is foolhardy. It typically produces overly complex systems with poor performance. Chapter 16 discusses the numerous ways you can preprocess data and why it is so important for proper data pattern recognition and price forecasting.

Chapter 17 discusses how to select the right input and target values for modeling your data. Modeling tools are compared, and a data-mining technology is chosen because it very quickly configures a models size and shape, automatically. To verily its effectiveness, an example trading system is produced and evaluated.

Chapter 18 shows how to build a trading system where technology is not used to forecast price action, but rather to detect useful price patterns, such as the head-and-shoulder pattern. Classical rules are then constructed to make trading decisions.

Developing a Trading System Using Intermarket Analysis

The study of a time series of financial data involves the analysis of a stream of market prices over time, in an effort to create a model that describes its behavior. Once the model is quantified, it can be utilized in the prediction of future values.

Intermarket analysis involves the study of a second data stream, to attempt to determine whether it can provide information that will assist in predicting the first.

The ultimate goal of the technical trader is to develop a system that has a high statistical probability of success. This can be accomplished by using readily available tools to improve the odds beyond random results.

In this chapter, we will explore the theoretical and practical aspects of intermarket analysis of time series data. We will use as an example, the development of a trading system that predicts the future value of the S&P 500 Index based on an analysis of the S&P itself, and on an intermarket analysis with the Treasury bond market.

A study of past S&P market prices alone can provide some information that allows us to forecast future prices with better than random accuracy. That market has exhibited a consistent trend over the years, and many simple technical analysis indicators, such as moving averages and oscillators, will have strong predictive value.

However, if an additional filter is imposed whereby trades in conformance with the prediction are only made if they are preceded by a signal from the T-bond market, much greater forecasting accuracy can be achieved.

In developing a trading system, careful attention must be paid to issues such as system complexity and data adequacy. Consideration of these factors improves the likelihood that the system will perform well in the future.

Developing an Approach

The field of statistics involves the numerical description of large sets of data. Beyond the description, one can create inferences about the data. For example, one can

determine the moving average of a time series of data. One can then infer that if prices are above the moving average, there is an upward trend; if prices are below the moving average, there is a downward trend.

The statistical analysis of a data stream allows us not only to extract useful information from the data, but also to determine the accuracy of the inferences we may derive from this data.

Statistical tools, such as linear regression, multiple regression, and statistical modeling, have been with us for a long time, and are used by many market analysts. However, their use requires substantial expertise in two fields: (1) technical expertise in the properties of the markets in question, and (2) expertise in the interpretation of multivariate statistics.

In using statistical tools, the market expert must first interpret market behavior. He or she must then analyze and manipulate the data, and preprocess it into a more meaningful form. Mistakes in any of these choices or processes will result in an invalid analysis.

Furthermore, a detailed knowledge of statistical analysis is required. Decisions must be made as to what type of model will adequately portray the data, and what analysis tools are to be used. Then, the results must be interpreted to determine their statistical significance.

With the use of modern technical analysis software, traders can now easily design technical trading systems, and optimize and backtest them. The programs are powerful tools that can complement financial knowledge and statistical expertise. They allow for easy interpretation of the value of a statistical model by measuring the bottom line, which is the profitability of a trading strategy based on the model.

modeling the data

The development of every trading system should start with a rational observation about market behavior. The software should pick out isolated profitable patterns from even random data, so care must be taken in developing and understanding relationships that are meaningful. If the significance of a particular pattern is understood, and can be verbalized, the pattern is more significant and more likely to be repeated into the future.

To try to predict future values in a time series, we use the principle of modeling. Simply stated, this involves trying to find variables that can describe the data, and then determining how the variables are related.

One common model of the value of a time series, such as a stock or commodity price, over time is:

Y1=T1 + C1 + S1 + R1