|
...analysis. There is considerable interest in using time series and data mining/ computational intelligence techniques to analyze the extensive historical datasets on stock prices. Such an analysis is made possible by the rapid advances made in computing power and information technologies. In order to improve the prediction accuracy one first needs to identify important indicators, and then use the appropriate prediction technique. Influential indicators can be determined using data mining techniques.
The Efficient Market Hypothesis (EMH) states that market prices follow a random walk model and cannot be predicted based on their past behavior (Fama, 1991). In addition to this, many factors that can affect the financial markets are interdependent, for example the general economic conditions, a trader's expectations, and political events. Therefore, stock market prediction is regarded as being a challenging task. Some researchers have questioned the validity of the random walk hypothesis. For example, Gencay (1998) found that technical trading rules are more successful at predicting exchange rates than a model that uses a random walk approach. Technical analysis can be used to explore anomalies in stock market behavior. These anomalies can be used to earn abnormal returns which violate the strong or semi-strong form of the EMH (Leigh et al., 2005).
Several studies examine the relationships between stock prices and fundamental variables on technical indicators. Some fundamental variables such as earning yield, cash flow yield, book-to-market ratio, and size have been shown to be able to predict stock returns (Banz and Bren, 1986; Jaffee et al., 1989; Fama and French, 1992; Chen et al., 2003). Moreover, macroeconomic variables such as short-term interest rates, expected inflation, dividend yields, and lagged returns have also been succesfully used to predict stock returns (Fama and Schwert, 1977; Campbell, 1987; Chen et al., 2003). Technical analysis is concerned with the dynamics of the market price and volume behavior for price prediction and ignores the EMH. Technical analysis patterns such as bull flag, moving averages, and volume spikes have been used to determine the stock price. The results of these studies show that these methods are either superior to or at least as good as the other methods (Gencay, 1998; Leigh et al., 2002; Leigh et al., 2005). Ausloos and Ivanova (2002) have investigated the relationship between momentum indicators and kinetic energy theory while Leus et al. (2001) have used genetic algorithms with technical indicators to analyze financial data.
The tendency in the field of financial forecasting is to use state variables that are fundamental or macroeconomic variables. Technical analysis, also known as charting, is widely used in real life and is based on using technical indicators to guide an investor on whether or not to buy or sell a stock. We try to understand the relationship between stock price and these indicators, which are computed using stock prices and trading volumes. More than 100 indicators have been developed to understand stock market behavior and thus the identification of the right indicators is a challenging problem. Two different approaches will be given in this paper. The first approach uses statistical techniques such as kernel Principal Component Analysis (kPCA) and factor analysis to identify the most important indicators. The second approach is based on heuristic models. Our goal is to compare the performances of these two selection approaches. We also compare two machine learning techniques, namely Support Vector Regression (SVR) and Neural Networks (NNs).
The SVR algorithm that was developed by Cortes and Vapnik (1995) and Vapnik (1995) is based on statistical learning theory. Using a regression approach, the goal is to construct a hyperplane that lies "close" to as many of the data points as possible (Cortes and Vapnik, 1995; Pontil and Verri, 1997; Ancona, 1999), thereby determining the trend line of the training data. Most of the points deviate with at most an [epsilon] precision from the target-defining [epsilon]-band. Therefore, the objective is to choose a hyperplane where w has a small norm (providing good generalization) while simultaneously minimizing the sum of the distances from the data points, outside the [epsilon]-band, to this hyperplane (Osuna et al., 1997; Smola and Scholkopf, 1998).
NN models have been applied to stock index and exchange rate forecasting (Hutchinson et al., 1994; Tsaih, 1999; Leigh et al., 2002). SVR has also been applied to stock price forecasting and option price prediction (Trafalis and Ince, 2002; Trafalis et al., 2003). Recent papers have showed that SVR outperforms Multi Level Perceptron (MLP) networks (Tay and Cao, 2002; Trafalis and Ince, 2002). This can be explained by noting that SVR involves a small number of free parameters and is a convex quadratic optimization problem. We will compare the results attained using SVR and MLP networks on short-term forecasting problems.
One of the goals of financial modeling is asset evaluation. The behavior of an asset can be analyzed by using either technical tools or parametric pricing methods, or by a combination of these methods. Since financial markets are complex, nonstationary, and deterministically chaotic systems, it is very difficult to forecast changes using deterministic (parametric) techniques because of the assumptions inherent in the parametric techniques: for example, linear regression models assume normality, serial correlation, etc. Therefore, nonparametric techniques such as SVR, NN, and time series models are good candidates for financial time series forecasting. NNs and radial basis function networks have been used to model a diverse range of financial situations including option pricing, stock index trading, and currency exchange rates (Hutchinson et al., 1994; Galindo, 1998; Leigh et al., 2002; Chen et al., 2003; Trafalis et al., 2003). NNs are universal function approximators that can map any nonlinear function without...
NOTE: All illustrations and photos
have been removed from this article.
More articles from IIE Transactions
Uncertainty management in optimal disassembly planning through learnin..., June 01, 2007
Identifying and visualizing nonlinear variation patterns in multivaria..., June 01, 2007
Looking for additional articles?
Search our database of over 3 million articles.
Looking for more in-depth information on this industry?
Search our complete database of Industry & Market reports by text, subject, publication
name or publication date.
About Goliath
Whether you're looking for sales prospects, competitive information, company
analysis or best practices in managing your organization,
Goliath can help you meet your business needs.
Our extensive business information databases empower business
professionals with both the breadth and depth of credible,
authoritative information they need to support their business
goals. Whether it be strategic planning, sales prospecting,
company research or defining management best practices -
Goliath is your leading source for accurate information.
|
