We observe that professionals and academics from each of technical analysis, quantitative analysis and fundamental analysis of the financial markets are beginning to recognize the importance of each other’s approach, and there is a tendency for the three to converge. The aim of this research is to combine technical analysis and quantitative analysis into a unified theory, which is the first major step in this convergence.
Another motivation comes from our real-data neural network experiments which indicate a significant performance of over 90% correctness in predicting the next-day index change direction. It turns out both necessary and feasible to develop a comprehensive computational theory of stock market behavior, which could predict the market direction at multiple time frames using only index or price time series data. The ideas presented in this paper constitute a basic framework of such a theory which we shall call the Swingtum theory in the following and future discussions.
Stock markets are complex dynamical systems whose elements are investors and traders with varying capital sizes using different investment and trading strategies. The behavior of the stock market of a given country is generally measured by a benchmark index such as ASX S&P 200 index for Australian stock market. The index is recorded in a time series whose time unit is usually one day or one minute. The daily index time series is the primary data source for technical analysis and quantitative analysis for short-term traders and mid- to long-term investors with time frames ranging from days through years to decades.
The minutely index time series is the main data source for day traders with a time frame ranging from a few minutes through hours to a few days, normally less than a week. Each data point contains usually 5 numbers: the open, high, low, close price and the traded volume of that time unit. In general, if we only consider one number, it is the close price. Still the time series of an index is considered a univariate time series, even the single variable may be a vector of 5 elements.
Looking inwards, a stock market has its internal structure consisting of a number of sectors, so the complete collection of all the sector indexes for a stock market may be called the intramarket time series which must be multivariate. Looking outwards, a stock market is positioned in the international stock market ecological systems. It may have its super markets that influence itself in an asymmetrical way, and there are also a number of other interrelated stock markets on the same level, so the influences are not uni-directional. Taking the Australian stock market as an example, the US stock market as represented by three indexes – Dow Jones, S&P 500, and NASDAQ may be considered its super market, while the Japanese, Hong Kong, Korean, German, French and British stock markets are obviously its neighbors on the same level.
The super and neighbor stock markets for a given stock market as well as typical bond markets, gold and oil markets and other interrelated financial markets may form the intermarkets of this stock market. The univariate time series of the stock market and the multivariate time series of its intramarkets and intermarkets provide the complete primary data source for prediction of the stock market movement. The secondary data source includes mainly the news (including all kinds of textual information) available from the Internet. However, this secondary data source is extremely irregular, and requires sophisticated techniques for automatic processing and interpretation.
This research aims at developing a comprehensive computational theory about the dynamics of the stock market for predicting the stock market index movement on multiple time frames from the primary data source of the index time series of this market, its intramarkets and its intermarkets, as well as from the automatic analysis of the news from Internet.
We shall call the complete collection of our views to the stock market, our assumptions, our mathematical models as well as computational procedures the Swingtum theory. Although this theory will keep evolving, our basic view to the market is fundamental. Our inspirations come from a century-old technical analysis and the recent progresses in quantitative analysis of the financial markets, including econometrics and econophysics.
A companion paper (Pan 2003) provides a joint review on technical and quantitative analysis and also has pointed out the possibility of a unified science of intelligent finance. More references on quantitative finance are provided by Sornette (2003), Farmer (1998), Farmer and Joshi (2002), Lo and McKinlay (1999), Campbell et al (1997), Mandelbrot (1982), Zhou and Sornette (2003). Comprehensive coverage on technical analysis can be found in Murphy (1999), Achelis (2000), Bulkowski (2002), and Prechter (2002).
The paper is organized as follows. Section 2 establishes our fundamental view to the stock market which is expressed in the Swing Market Hypothesis. Section 3 formulates a simple but general dynamic model of the market returns. Sections 4 and 5 presents the parametric models for characterizing fractal and cyclical market movements. Section 6 describes the quantum price-time space in which the market is supposed to travel as driven by endogenous dynamics and exogenous news shocks or impacts. Section 7 outlines a computational approach of nonparametric nearest neighbor pattern recognition exploiting multidimensional chaos in price time series. Finally section 8 concludes the paper.
Prof. Heping Pan
Next: The Swing Market Hypothesis
Summary: Index