Ebook Financial News Analysis for Intelligent Portfolio Management
An important premise in financial investing is that there must be a reasonable amount of validated information before a security is considered from an investment standpoint. Given the requirements of having various expertise and the difficulties in locating and evaluating information sources, financial portfolio management has to date carried out by investment firms that employ teams of specialists for finding, filtering and evaluating relevant information. It has primarily focused on the portfolio selection process (i.e., asset allocation) as opposed to portfolio monitoring the ongoing, continuous, daily provision of an up-to-date financial picture of an existing portfolio.
In portfolio management, it is important for an investor to monitor his or her portfolio regularly in addition to asset allocation, because it must be determined whether or not the return results of the portfolio meet the expectations of the investor, or whether there is a need to change the strategic asset allocation. The monitoring process also provides comprehensive, detailed information on the investment positions of the investor. The result of the controlling monitoring might require changes in the asset allocation in order to realign the long-term asset allocation strategy. It is important to note that portfolio management, as an investment process, is not a static, but a dynamic one, where you should regularly adapt your decisions to changes in the market and in your own circumstances.
In the application domain of portfolio management, a large volume of information exists about a company and its financial performance that humans must effectively attend to and manage in order to make decisions. To address this problem, we proposed and implemented a multi-agent system, called Warren. Warren is composed of several agents that help the user manage his or her portfolio by providing quantitative information: stock price, performance history, earnings summaries and risk (¬ value), and to proactively advise the user whenever the portfolio may be too risky for the user’s specified tolerance to risk.
In addition to such quantitative information, it is desirable to look into qualitative data such as financial news reports, in order to get multiple perspectives on the financial performance of the company of interest, because there is a positive correlation between news reports on a company’s financial outlook and its attractiveness as an investment. However, because of the tremendous volume of such reports, it is impossible for financial analysts or investors to track and read each one. Therefore, it would be very helpful to have a technology for automatically analyzing news reports that reflect positively or negatively on a company’s financial outlook. To accomplish this task, we devised and implemented a new agent, called TextMiner, which performs the tasks of information retrieval for news from on-line news providers such as Reuters, CNN Financial Network, Business Wire, Forbes.com and others.
The goal of TextMiner is to provide an accounting of news articles on the company of interest for a period, in terms of good or bad financial performance. As a software agent in Warren, the TextMiner agent, upon a request from the user or other agents, selectively attends to news reports on the company of interest by filtering non-financial news out and then classifying them in terms of the company’s current financial status.
We devised a new text classification method that helps TextMiner carry out its classification task. The devised method predicts the class of a financial news article through the voting process among experts, which are frequently co-located phrases. A co-located phrase is a sequence of nearby but not necessarily consecutive words. In addition, it is important for an agent to learn the content-shift autonomously, because the vocabularies of text domains change slightly from time to time, and the intervention by humans in order to label text data is quite expensive. The devised method for providing TextMiner with self-learning capability estimates the class of unlabeled data on the basis of the learner’s confidence, which is obtained through the training phase.
In this paper, we present Warren, which is a multi-agent system for intelligent portfolio management, motivated by the great benefits of working in teams within the domain of Distributed Artificial Intelligence (DAI), and TextMiner, which is a text classification agent that takes advantage of information retrieval techniques to complement quantitative financial information.
The paper is organized as follows. Section 2 details characteristics of portfolio management domain and our previous approaches to this domain. Section 3 describes text analysis for augmenting management of portfolio. Section 4 takes an example of intelligent portfolio management. Section 5 discuss the results and future works.
Contents
1 Introduction
2 Portfolio Management Domain
- 2.1 Warren: A Multi-Agent System for Intelligent Portfolio Management
3 TextMiner: An Agent for Text Analysis
- 3.1 InformationRetrievalTasksofTextMiner
- 3.1.1 Segmentationof News Articles
3.1.2 Classification of News Articles
3.1.3 Evaluation of Financial News Classification
4 Putting It All Together
5 Conclusion and Future Work
6 Acknowledgements
Download
PDF Ebook Financial News Analysis for Intelligent Portfolio Management
Posted in :
