Benoit Mandelbrot was amongst the first to notice volatility clustering

Workpackage MS Measuring Sentiment in the Market [Ahmad, Kearney]

This research question posed in this workpackage is this: to what news about the markets and the broader environment effects stock returns that exhibit momentum? Specifically, our investigation will focus on developing methods and techniques for studying how causes of volatility, as instances of bounded rationality and contrarian behaviour, manifests itself in news reports, official statements and blogs. The research question posed here is of considerable import for financial economics and mathematics because there is some evidence that ‘linguistic media content captures otherwise hard-to-quantify aspects of firms’ fundamentals, which investors quickly incorporate in stock prices’ (Tetlock 2007). Equally importantly the research question is relevant for the burgeoning fields of sentiment analysis and multi-media information extraction systems. For the former, which is reported increasingly in the literature on finance, the choice of the method of linguistic analysis, and the data used therein, appears arbitrary and there is a need to ground the analysis, and the choice of data sets, in a transparent formal framework (Picard 2000, Pavia, Prada and Picard 2007, and Ahmad 2008). For multi-media information extraction systems, this fusion of numbers and texts, together with graphics, as a precursor for decision making will help in a well-grounded development of such systems for neuroscience research (Driver and Spence 2005), for systems that learn to fuse data (Koppel and Shtrimberg 2004), and for annotating images with words and vice versa (Barnard et al 2003, Ahmad 2003).

Benoit Mandelbrot was amongst the first to notice volatility clustering in the diachronic behaviour of financial instruments(1963); latterly, Engle explained that the clustering is largely due to information arrival (1993, 2003). Typically in econometrics and financial economics, information proxies are used (Poon and Granger 2003, and Wang, Keswani and Taylor 2004). The information arrival can be a pure count of the number of news announcements (Engle and Ng 1993), or intraday exchange rate fluctuations (Andersen et al 2002, Chang and Taylor 2003), or stock prices (Liden 2006), or in terms of firm-level proxies, including closed-end fund discount, dividend premium and so on (Baker and Wurgler 2006) or in terms of a firm’s size and momentum strategies (Hong, Lim and Stein, 2000). There has been some attempt to look at the content of news stories rather on the proxies about a decade ago when a keyword analysis was carried out of public information releases relating to foreign exchange markets: the frequency of keywords, related to the names of firms and instruments, and the frequency of an arbitrary set of affect (positive, negative or neutral or no-affect) was computed and correlated with the price movement (DeGennaro and Shreve 1997). There are a number of publications in the computing, especially information extraction literature where we find linguistic analysis performed at higher level of linguistic description – complex thesaurus of affect words (Wiebe, Wilson, and Cardie 2005), and general-purpose, electronic thesaurus, like WordNet (developed by Miller and Fellbaum 1995) together with the analysis of syntax that are used to encode sentiment (Surudeanu et al 2003, Chan and Lam 2005, Lan et al 2005, Kennedy and Inkpen 2006). Financial news has been analysed at the level of vocabulary and meaning for automatically extracting sentiment bearing patterns in English, Arabic and Chinese (Ahmad, Cheng & Almas 2006) without recourse to an arbitrarily selected thesaurus (Almas & Ahmad 2006). Studies of how sentiments are articulated differently in everyday language and in the language of finance will throw some light on how stakeholders choose to influence the market (Devitt and Ahmad 2007)

Learning to identify ‘sentiments’ in financial texts has been undertaken using neural networks (Koppel and Shtrimberg 2004) and statistical classifiers (Das and Chen 2006). The work described in finance and econometrics focuses on information proxies and treats time variation of prices and volumes using sophisticated regressive models like GARCH. The work in computing treats language more carefully but lacks the rigour when it comes to analysing the time series of prices and volumes.

The use of carefully constructed affect dictionaries to understand the contents of arbitrary texts is well rooted in content analysis pioneered by the political scientist Harold Laswell (1948) who had analysed changes in political and social sentiments leading to landmark events (coalescing of Democrat and Republican policies and the onset of the 2^nd World War, Namenwirth & Laswell 1970); latterly Stone et al (1966) used such analysis in a range of socio-economic analysis (Kelly and Stone 1975). Recently, Tetlock (2007), using Sone and Laswell’s Dictionary of Affect, for analysing opinions expressed in Wall Street Journal, together with a sophisticated GARCH model to show the influence of ‘bad news’ on historical volatility. Ahmad has used more diverse text over a 10 year period to show that the terms with negative polarity in themselves have higher volatility than the positive polarity (Ahmad 2008). Indeed, the boom in commercially available systems, developed by Reuters News Service and Dow-Jones Financial Services, now offer a sentiment analysis service based on an arbitrarily selected thesaurus of affect words.
Outputs and Timetable: Workpackage MS will lead to the development of well-grounded methods and techniques, in information extraction and in financial economics, for analysing the effect of news on volatility in general. The workpackage requires one research assistant (RA) and two PhD students.
Year 1: Evaluate existing sentiment analysis programs and associated data sets (prices/volumes and thesauri) as used in finance studies, in computing science, and sentiment analysis programs available commercially. PhD student 1 will work on the effect of news on ‘slow’ volatility clustering data and PhD student 2 will work on the effect of news on ’medium’ duration volatility clusters. Research Assistant to work on the impact of news of high-frequency volatility clusters in close conjunction with momentum transfer workpackage. Create a ‘sentiment’ Roundtable comprising workers in financial services industry in Dublin. Participate in learned conferences and workshops.
Year 2: Build prototype system for sentiment analysis as specified by the Roundtable with sub-systems for slow, medium duration and high frequency volatility clusters. Incorporate learning algorithms developed in the system (Input from and dependence on WP on soft-computing). Create an evaluation strategy
Year 3: Test the system that will be able to take real-time data and will offer a choice of affect dictionaries or the automatic extraction of sentiment. Disseminate work widely through peer-reviewed publications, popular science press, and technical journals.
References

Ahmad, K. Notes on Learning to Compute and Computing to Learn. (Invited Keynote Speech). International Conference on Machine Learning and Applications (ICMLA'03) Los Angeles, California, June 23-24, 2003. pp 107-117.

Ahmad, K., D. Cheng and Y. Almas. (2006) ‘Multi-lingual Sentiment Analysis of Financial News Streams.’ In (Eds.) Stefano Cozzini, Stefano d'Addona and Rosario Mantegna. Proc. 1st Int. Conf. on Grid in Finance (Palermo, February 2006) (http://pos.sissa.it/archive/conferences/026/001/GRID2006_001.pdf).

Ajith Abraham and Crina Grosan, Engineering Evolutionary Intelligent Systems: Methodologies, Architectures and Reviews, Engineering Evolutionary Intelligent Systems, Studies in Computational Intelligence, Springer Verlag, Germany, ISBN 978-3-540-75395-7, pp. 1-22, 2008.

Ajith Abraham, Swagatam Das and Sandip Roy, Swarm Intelligence Algorithms for Data Clustering, Soft Computing for Knowledge Discovery and Data Mining, Oded Maimon and Lior Rokach (Eds.), Springer Verlag, Germany, ISBN 978-0-387-69934-9, pp. 279-313, 2007.

Almas, Y., & K. Ahmad. (2006).‘LoLo: A System based on Terminology for Multilingual Information Extraction’. In (Eds.) Mary Elaine Califf, Mark A. Greenwood, Mark Stevenson & Roman Yangraber. COLING ACL 2006: Workshop on Information Extraction beyond the Document, Sydney, Australia, 22 July 2006. Association of Computational Linguistics. pp56 – 65. (http://acl.ldc.upenn.edu/W/W06/W06-0207.pdf)

Almas, Yousif., and Khurshid Ahmad. (2007). A note on extracting ‘sentiments’ in financial news in English, Arabic & Urdu, The Second Workshop on Computation, al Approaches to Arabic Script-based Languages, Linguistic Society of America 2007 Linguistic Institute, Stanford University, Stanford, California., July 21-22, 2007, Linguistic Society of America. pp 1-12.

Andersen, T. G., Bollerslev, T., Diebold, F X., & Vega, C. (2002). Micro effects of macro announcements: Real time price discovery in foreign exchange. National Bureau of Economic Research Working Paper 8959, http://www.nber.org/papers/w8959

Baker, Malcolm., & Wurgler, Jeffrey (2006). Investor Sentiment and the Cross-Section of Stock Returns, Journal of Finance, Vol 59, pp 1645-1680

Barnard, K., Duygulu, P., Forsyth, D., de Freitas, N., Blei, D.M., and Jordan, M.L., Matching Words with Pictures, Journal of Machine Learning Research, Vol. 3, pp 1107-1135, 2003.

Bauwens, L.; W.B. Omrane; and P. Giot. (2005). “News Announcements, Market Activity and Volatility in the Euro/Dollar Foreign Exchange Market.” Journal of International Money and Finance. Vol. 24, pp 1108-1125.

Chan, K. and Lam, W. (2005) ‘Extracting Causation Knowledge from Natural Language Texts’. International Journal of Intelligent Systems, 20(3), pp. 327-358.

Chang, Y., and Taylor, S.J. (2003) Information Arrivals and Intraday Exchange Rate Volatility. Journal of International Financial Markets, Institutions and Money, vol 13, pp85-112

Das, S.R and Chen, M. Y. (2006) Yahoo! for Amazon: Sentiment Extraction From Small Talk on the Web. Working Paper, Leavey School of Business, Santa Clara University, California. (Available at: http://scumis.scu.edu/~srdas/chat.pdf).

DeGennaro, R., and R. Shrieves (1997): “Public information releases, private information arrival and volatility in the foreign exchange market” . Journal of Empirical Finance Vol 4, pp 295–315.

Devitt, Ann., and Ahmad, Khurshid. ‘Sentiment Polarity Identification in Financial News: A Cohesion-based Approach’. Proceedings of the 45th Annual Meeting of the Association for Computational Linguistics. June 23–30, 2007, Prague, Czech Republic. Stroudsburg, PA: Association for Computational Linguistics (ACL) pp 984-991.

Engle, Robert F. (2003). Econometric models and financial practice. http://nobelprize.org/nobel_prizes/economics/laureates/2003/engle-lecture.html

Engle, R. F. and Ng, V. K (1993). Measuring and testing the impact of news on volatility, Journal of Finance Vol. 48, pp 1749—1777.

Hong, Harrison., Terence Lim., and Jeremy Stein. (2000). Bad News travels slowly: Size, Analysts Coverage and the Profitability of Momentum Strategies. Journal of Finance. Vol 55, (No. 1), pp 265-295.

Kelly, Edward., and Stone, Philip. (1975). Computer Recognition of English Word Senses. Amsterdam: North-Holland Linguistic Series

Kennedy, Alistair & Inkpen, Diana. (2006). Sentiment classification using contextual valence shifters. Computational Intelligence . Vol. 22 (No. 2), pp 117-125.

http://en.wikipedia.org/wiki/Credit_crunch (Accessed 17th April 2008)

Koppel, M. and Shtrimberg, I. (2004) ‘Good News or Bad News? Let the Market Decide’. In AAAI Spring Symposium on Exploring Attitude and Affect in Text, Palo Alto: AAAI Press, pp. 86-88.

Lan, KC., Ho, KS., Pong Luk, RW. and Yeung, D.S. (2005) FNDS: A Dialogue-based System for Accessing Digested Financial News. Journal of Systems and Software, 78(2), pp. 180–193.

Lasswell, Harold D. (1948). Power and Personality. London: Chapman & Hall.

Lidén, Erik R. (2006) 'Stock Recommendations in Swedish Printed Media: Leading or Misleading?', The European Journal of Finance, Vol 12 (No.8), 731-748

Mandelbrot, Benoit. (1963). ‘The variation of certain speculative prices’. Journal of Business. Vol. 36, pp 394-419.

Namenwirth, Zvi., and Lasswell, Harold D. (1970) The changing language of American values : a computer study of selected party platforms. Beverly Hills (Calif.) : Sage Publications.

Paiva, Ana., Prada, Rui., Picard, Rosalind W. (2007) (Editors). Proceedings of Affective Computing and Intelligent Interaction: Second International Conference, ACII 2007, Lisbon, Portugal, September 12-14, 2007. Berlin and Heidelberg: Springer Verlag.

Paul C. Tetlock , Saar-Tsechansky, Mytal, and Mackskassy, Sofus (2005). More Than Words: Quantifying Language to Measure Firms’ Fundamentals. (http://www.mccombs.utexas.edu/faculty/Paul.Tetlock/papers/TSM_More_Than_Words_09_06.pdf)

Peramunetilleke, D. and Wong, R. K. (2002) ‘Currency Exchange Rate Forecasting from News Headlines’. In Proc. of the Thirteenth Australasian Database Conference ADC2002, Melbourne, Australia, pp. 131-139.

Picard, Rosalind. (2000). Affective Computing - 1st edition. Boston: The MIT Press.

Poon, S., & Granger, C.W.J. (2003). Forecasting financial market volatility: a review, Journal of Economic Literature, 41, 478-539

Pui, G., Fung, C., Yu, J. and Lu, H (2005) The Predicting Power of Textual Information on Financial Markets. IEEE Intelligent Informatics Bulletin, 5(1), pp. 1-10.

Robert F. Engle III (2003). Risk and Volatility: Econometric models and Financial Practice. – A Nobel Lecture, December 8, 2003

Seo, Y., Giampapa, J. and Sycara, K. (2004) ‘Financial News Analysis for Intelligent Portfolio Management’. Technical Report CMU-RI-TR-04-04, Robotics Institute, Carnegie Mellon University, Pennsylvania, USA. (Available at: http://www.ri.cmu.edu/pubs/pub_4570.html).

Stone, Philip J.,Dexter C. Dunphy, Marshall S. Smith, Daniel M. Ogilvie et al (1966). The General Inquirer: A Computer Approach to Content Analysis. Boston:The MIT Press.

Surdeanu, M., Harabagiu, S., Williams, J. and Aarseth, P. (2003) ‘Using Predicate-Argument Structures for Information Extraction’. In Proc. of the 41^st Annual Conference of the Association for Computational Linguistics (ACL’03), pp 8–15.

Tetlock, P. C. (2007). Giving content to investor sentiment: The role of media in the stock market. Journal of Finance, Vol. 62 (3), pp 1139-1168.

Wang, Yaw-Huei., Keswani, Aneel., and Taylor, Stephen J. (2004). "The Relationships between Sentiment, Returns and Volatility" . EFMA 2004 Basel Meetings Paper Available at SSRN: http://ssrn.com/abstract=499502.

Wiebe, Janyce., Theresa Wilson, and Claire Cardie. (2005). Annotating expressions of opinions and emotions in language. Language Resources and Evaluation. Vol 39(2-3), pp 165–210.

Surdeanu, M., Harabagiu, S., Williams, J. and Aarseth, P. (2003) ‘Using Predicate-Argument Structures for Information Extraction’. In Proc. of the 41^st Annual Conference of the Association for Computational Linguistics (ACL’03), pp 8–15.

Tetlock, P. C. (2007). Giving content to investor sentiment: The role of media in the stock market. Journal of Finance, Vol. 62 (3), pp 1139-1168.