SEARCH
ISTC-CC NEWSLETTER
RESEARCH HIGHLIGHTS
Ling Liu's SC13 paper "Large Graph Processing Without the Overhead" featured by HPCwire.
ISTC-CC provides a listing of useful benchmarks for cloud computing.
Another list highlighting Open Source Software Releases.
Second GraphLab workshop should be even bigger than the first! GraphLab is a new programming framework for graph-style data analytics.
ISTC-CC Abstract
StockMarket Volatility Prediction: A Service-Oriented Multi-Kernel Learning Approach
Proceedings of IEEE Int. Conf on Service Computing (SCC'12), June 2012.
Feng Wang, Ling Liu*, and Chenxiao Dou
Wuhan University
*Georgia Institute of Technology
Stock market is an important and active part of nowadays financial markets. Stock time series volatility analysis is regarded as one of the most challenging time series forecasting due to the hard-to-predict volatility observed in worldwide stock markets. In this paper we argue that the stock market state is dynamic and invisible but it will be influenced by some visible stock market information. Existing research on financial time series analysis and stock market volatility prediction can be classified into two categories: in depth study of one market factor on the stock market volatility prediction or prediction by combining historical price fluctuations with either trading volume or news. In this paper we present a service-oriented multi-kernel based learning framework (MKL) for stock volatility analysis. Our MKL service framework promotes a two-tier learning architecture. In the top tier, we develop a suite of data preparation and data transformation techniques to provide a source-specific modeling, which transforms and normalizes a source specific input dataset into the MKL ready data representation. Then we apply data alignment techniques to prepare the datasets from multiple information sources based on the classification model we choose for cross-source correlation analysis. In the next tier, we develop model integration methods to perform three analytic tasks: (i) building one sub-kernel per source, (ii) learning and tuning the weights for sub-kernels through weight adjustment methods and (iii) performing multi-kernel based cross-correlation analysis of market volatility. To validate the effectiveness of our service oriented MKL approach, we performed experiments on HKEx 2001 stock market datasets with three important market information sources: historical prices, trading volumes and stock related news articles. Our experiments show that 1) multi-kernel learning method has a higher degree of accuracy and a lower degree of false prediction, compared to existing single kernel methods; and 2) integrating both news and trading volume data with historical stock price information can significantly improve the effectiveness of stock market volatility prediction, compared to many existing prediction methods.
KEYWORDS: multiple kernel learning; stock prediction; support vector machine; multi-data source integration;
FULL PAPER: pdf