Chapter Preview
TopBackground
A human trader’s capability to digest financial news “on the fly” is limited in speed and in quantity. However, with increasing computational capacities and faster text mining methods nowadays, new automated news reading and interpreting technologies are available to efficiently extract, aggregate and categorise large volumes of information streams in an instant.
Text mining refers to the application of artificial intelligence to automatically extract information from machine-readable sources (XML, HTML, JSON, and most other electronic documents) by distinguishing and detecting linguistic patterns in written text. Using the accuracy and efficiency of a computer, text mining tools aim to mimic a human’s ability to comprehend contextual meaning by looking for repeated patterns, key terms, named entities, or similar subsets thereof in a vast collection of texts within seconds.
In recent years, more and more market participants consider the addition of financial news analytics into their algorithmic trading engine to better predict the direction or volatility of market movements before making an investment decision. Particularly in high-frequency trading, these decisions need to be made almost immediately to reduce latency. Financial news analytics, however, is not just limited to textual data mining exercises. It is a new interdisciplinary research area requiring knowledge and expertise from computer science, finance, and economics. The sole implementation of advanced machine-learning algorithms without the consideration of market microstructure effects and economics of financial markets will have only little value for investors in real-world applications.
Financial news analytics combines methods from information retrieval, statistical learning, natural language processing, and financial econometrics to collect, categorise, interpret unstructured textual input data and convert this into metric output data, such as a financial sentiment score. In the following, we first give an overview of the different common information protocols used in financial applications. We then discuss the implementation of news analytics in an investment strategy in five steps (see also Johnson, 2010): news filtering (what is economically or financially relevant?), news association (what is interesting for which investor?), news interpretation (what does the news mean?), econometric modelling (how can the return and its volatility be statistically modelled?), and strategy testing and implementation (how can the trader capitalise the information?).
Key Terms in this Chapter
Text Data Mining (or Text Analytics): Is the process of extracting high-quality information from textual data (i.e. non-numeric data). It usually involves parsing the input text, identifying linguistic features, extracting patterns within the structured data. Common methods applied in text mining are, for example, text categorization, text clustering, or document summarization.
High-Frequency Trading (HFT): Is a special form of electronic trading characterised by the use of complex computer algorithms to analyse quote data, trigger readily implemented trading strategies, and reduce trade execution latency for a fast reallocation or turnover of trading capital within fractions of a second.
Natural Language Processing (NLP): Is a specialised research area in computer science focusing on algorithms enabling computers to derive meaning from human (natural) languages. Typical tasks in NLP are, for example, word and sentence tokenization, information extraction, text classification, etc.
Macroeconomic News: Refers to pre-scheduled news releases of macroeconomic indicators (e.g., gross domestic product, inflation rate, central bank interest rates, job market statistics, etc.) by public institutions with high credibility ratings (e.g. US Federal Reserve, Central Banks, etc.).
Sentiment Analysis: Is the application of natural language processing and computational linguistics to analyse textual data with the aim to associate the (individual or collective) opinions of authors towards a certain topic. In its simplest form, it focuses on sentences that signal positive, negative, or neutral attitudes and then derives a numerical statistic from the pattern matching in order to “quantitatively summarise” the written content.
Information Transmission Protocols: Refer to Internet protocol (IP) suite communication standards used for information exchange in computer networks.
Financial News Analytics: Combines methods from information retrieval, statistical learning, natural language processing, and financial econometrics to collect, categorise, interpret unstructured textual financial market data and convert this into metric output data, such as a financial sentiment score.