Movement of stocks in the financial market is a typical example of financial time series data. It is generally believed that past performance of a stock can indicate its future trend and so stock trend analysis is a popular activity in the financial community. In this chapter, we will explore the unique characteristics of financial time series data mining. Financial time series analysis came into being recently. Though the world’s first stock exchange was established in the 18th century, stock trend analysis began only in the late 20th century. According to Tay et al. (2003) analysis of financial time series has been formally addressed only since 1980s. It is believed that financial time series data can speak for itself. By analyzing the data, one can understand the volatility, seasonal effects, liquidity, and price response and hence predict the movement of a stock. For example, the continuous downward movement of the S&P index during a short period of time allows investors to anticipate that majority of stocks will go down in immediate future. On the other hand, a sharp increase in interest rate makes investors speculate that a decrease in overall bond price will occur. Such conclusions can only be drawn after a detailed analysis of the historic stock data. There are many charts and figures related to stock index movements, change of exchange rates, and variations of bond prices, which can be encountered everyday. An example of such a financial time series data is shown in Figure 1. It is generally believed that through data analysis, analysts can exploit the temporal dependencies both in the deterministic (regression) and the stochastic (error) components of a model and can come up with better prediction models for future stock prices (Congdon, 2003).
TopIntroduction
Movement of stocks in the financial market is a typical example of financial time series data. It is generally believed that past performance of a stock can indicate its future trend and so stock trend analysis is a popular activity in the financial community. In this chapter, we will explore the unique characteristics of financial time series data mining. Financial time series analysis came into being recently. Though the world’s first stock exchange was established in the 18th century, stock trend analysis began only in the late 20th century. According to Tay et al. (2003) analysis of financial time series has been formally addressed only since 1980s.
It is believed that financial time series data can speak for itself. By analyzing the data, one can understand the volatility, seasonal effects, liquidity, and price response and hence predict the movement of a stock. For example, the continuous downward movement of the S&P index during a short period of time allows investors to anticipate that majority of stocks will go down in immediate future. On the other hand, a sharp increase in interest rate makes investors speculate that a decrease in overall bond price will occur. Such conclusions can only be drawn after a detailed analysis of the historic stock data. There are many charts and figures related to stock index movements, change of exchange rates, and variations of bond prices, which can be encountered everyday. An example of such a financial time series data is shown in Figure 1. It is generally believed that through data analysis, analysts can exploit the temporal dependencies both in the deterministic (regression) and the stochastic (error) components of a model and can come up with better prediction models for future stock prices (Congdon, 2003).
Figure 1. A typical movement of a stock index
TopBackground
Financial time series are a sequence of financial data obtained in a fixed period of time. In the past, due to technological limitations, data was recorded on a weekly basis. Nowadays, data can be gathered for very short durations of time. Therefore, this data is also called high frequency data or tick by tick data. Financial time series data can be decomposed into several components. Kovalerchuk and Vityaev (2005) defined financial time series data as the summation of long term trends, cyclical variations, seasonal variations, and irregular movements. These special components make financial time series data different from other statistical data like population census that represents the growth trends in the population.
In order to analyze complicated financial time series data, it is necessary to adopt data mining techniques. Currently, the commonly used data mining techniques are either statistics based or machine learning based. Table 1 compares the two types of techniques.
Table 1. Comparison of statistical and machine learning techniques
| Statistical techniques | Machine learning techniques |
| Advantages | • Relatively simple to use
• Uni-variate and multi-variate analysis enable use of stationary and non-stationary models
• Users can select different models to fit data and estimate parameters of models | • Easy to build model based on existing data
• Computation carried out in parallel to model building which allows real time operations
• Able to create own information representation during learning stages
• More tolerant to noise in data |
| Disadvantages | • Performance and accuracy are negatively influenced by noise and non-linear components
• The assumption of repeat patterns is unrealistic and may cause large errors in prediction | • Unstable for very large problems
• Black box functions often do not provide any explanation of derived results |