You are here: Home › Finance Research Group

Finance Research Group

 

Simon AlfanoDr. Simon Alfano drummer.jpgDaniel Drummer Stefan FeuerriegelDr. Stefan Feuerriegel
Research Group Leader		 McK_Maerkle.jpgDr. Joscha Märkle-Huß Dirk NeumannProf. Dr. Dirk Neumann
Chair Holder

 

 

Research

Machine learning and text mining have been used for some time to forecast stock market reactions. However, knowledge on information processing by human agents facing qualitative information is rare and mostly unknown. The Finance Research Group aims to close this gap and improve the understanding of the role of textual financial news in forming investor decisions.

Spin-off: TonalityTech GmbH

Our spin-off TonalityTech GmbH provides an easy-to-use add-in for Microsoft Word for corporate communications, in particular investor relations and media relations. Our research revealed that an improved tonality of corporate communications can increase company valuation by up to 2.7%.

Our add-in makes this knowledge available to corporate communications practitioners. Thus, corporate communications practitioners can benefit from an evidence-based tool that facilitates steering the tonality and readability of corporate communication. At the same time, the add-in seamlessly integrates into existing communication creation processes.

Interested? Contact us via tonalitydumdummymy@is.dumdummymyuni-freiburg.de or visit our website.
 

Podcast and article by IR Magazine featuring our research

Recently, IR Magazine featured our research in a podcast in the series IR magazine asks. An article on our research is also available from IR Magazine.

 

Organization of FinanceCom 2016 workshop

The Chair for Information Systems Research will host the 2016 FinanceCom workshop in Frankfurt. After very successful FinanceCom workshops in Sydney, Regensburg, Montreal, Paris, Frankfurt and Barcelona, FinanceCom 2016 will be returning to Frankfurt on 8th December 2016. Advancements in Information and Communication Technologies have paved the way to new business models, markets, networks, services, and players in the financial services industry.

FinanceCom 2016 invites papers that help to understand, drive and exploit the associated systems, technologies and opportunities.

http://www.financecom2016.is.uni-freiburg.de

 

Project: Dictionary Generation for Financial News

 

Aims
  • Understanding how companies frame the language of their news disclosures
  • Analyzing information processing of investors in financial news 
  • Generating domain-specific dictionaries for sentiment analysis

 

Method
  • Data basis: Regulated ad hoc announcements language of German firms in English
  • Using stock market reactions as an objective measure to quantify news tone
  • Utilizing Bayesian variable selection methods to select decisive positive and negative key words

 

Results
  • Generated dictionaries using regularization methods outperform static dictionaries in terms of goodness-of-fit and predictive performance
  • Manually-selected word lists lead to misclassifications in a financial context
  • Investors expose the framing of positive into negative words when judging an investment

Kernel Density Estimation of generated dictionaries           Top 5 positive and negative words

 

Pröllochs N, Feuerriegel S, Neumann D: Generating Domain-Specific Dictionaries Using Bayesian Learning, 23rd European Conference on Information Systems (ECIS 2015), Münster, Germany, May 26-29, 2015.

Pröllochs N, Feuerriegel S, Neumann D: Say it Right – How Managers Prettify Corporate Disclosures 2015 (Winter Conference on Business Intelligence (WCBI 2015), Snowbird, Utah, 12-14 March 2015.

 

Pröllochs N, Feuerriegel S, Neumann D: Is Human Information Processing Affected by Emotional and Cognitive Biases? Evidence from the Stock Market, Working Paper, University of Freiburg, 2015.

 

Project: Information Processing of Initial Public Offering Filings

 

Aims
  • Stakeholders in Initial Public Offerings (IPO) must understand how filings affect pricing
  • Additional focus on the textual content of pre-IPO news coverage
  • Hypothesis that news and filings receive different attention

 

Method
  • Data basis for U.S. IPOs: Form 424 filings and Thomson Reuters News Archive for Machine Readable News 
  • Utilizing sentiment analysis to extract subjective information from filings and news
  • Linking sentiment and first-day returns to study news reception

 

Results
  • Degree of specific investor attention depends on the specific type of information
  • The more negative the pre-IPO news coverage, the higher the first-day returns
  • Risks are recognized in filings, whereas emphasis on company opportunities in news

News Reception around Initial Public Offerings

  • Conclusion: Information processing depends on the attention different information receives

 

Feuerriegel S, Schmitz J T, Pröllochs N, Neumann D: What Matters Most? How Tone in Initial Public Offering Filings and Pre-IPO News Influences Stock Market Returns What Matters Most? How Tone in Initial Public Offering Filings and Pre-IPO News Influences Stock Market Returns 2015 (2015 FMA European Conference, Venice, Italy, June 11-12, 2015.

Feuerriegel S, Schmitz J T, Neumann D: What Matters Most? How Tone in Initial Public Offering Filings and Pre-IPO News Influences Stock Market Returns, Working Paper, University of Freiburg, 2014.

 

Project: News Reception Compared Across Extracted Topics

 

Aims
  • Listed companies are typically obliged to publicly disclose any information that might affect their stock prices
  • Analyzing the effect of topics in such press releases on the stock market returns

 

Method
  • Data basis: Regulated ad hoc announcements of German firms in English
  • Utilizing sentiment analysis to extract  subjective information from filings and news
  • Linking sentiment and abnormal returns to study news reception

 

Results
  • Extracted 40 topics by utilizing the Latent Dirichlet Allocation
  • Median (abnormal) returns and news reception by investors varies greatly across different topics
  • News from many topics have no significant influence on stock perfomances, e.g. news covering legal, renewable energy or capital inceases
  • For example, M&A news and share issueing are negatively linked, whereas manager appointments and financial reports are positively received by investors at a significant level

Topic-based news reception

 

Feuerriegel S, Ratku A, Neumann D: Which News Disclosures Matter? News Reception Compared Across Topics Extracted from the Latent Dirichlet Allocation, Working Paper, University of Freiburg, 2015.

Feuerriegel S, Ratku A, Neumann D: Analysis of How Underlying Topics in Financial News Affect Stock Prices Using Latent Dirichlet Allocation, Working Paper, University of Freiburg, 2014.

 

Project: Speculation & Irrational Exuberance in Oil Markets

 

Aims
  • Understanding information processing oil markets
  • Finding evidence of speculation and irrational exuberance

 

Method
  • Data basis: Thomson Reuters News Archive for Machine Readable News 
  • Utilizing sentiment analysis to extract subjective information from news
  • Linking sentiment to abnormal returns and the economic cycle to study news reception

 

Results
  • News sentiment is a major driver of oil prices, more important than all fundamental variables in the comparison
  • We account for the endogeneity problem by performing an instrumental variable regression to show that news sentiment affects crude oil prices causally

news_reception

  • Evidence for exaggerated news reception as it comes along with a feedback loop (i.e. irrational exuberance)

 

Jandl J, Feuerriegel S, Neumann D: Long- and Short-Term Impact of News Messages on House Prices: A Comparative Study of Spain and the United States, 2014 (35th International Conference on Information Systems (ICIS 2014), Auckland, New Zealand, 14-17 December 2014, Completed Research Paper).

Feuerriegel S, Neumann D: News or Noise? How News Drives Commodity Prices, 2013 (34th International Conference on Information Systems (ICIS 2013), Milan, Italy, 15-18 December 2013, Completed Research Paper).

Feuerriegel S, Lampe M W, Neumann D: News Processing during Speculative Bubbles: Evidence from the Oil Market, 2014 (47th Hawaii International Conference on System Sciences (HICSS), Waikoloa, Big Island, January 6-9, 2014, IEEE Computer Society).

Feuerriegel S, Ratku A, Neumann D: Finding Evidence of Irrational Exuberance in the Oil Market, Working Paper, University of Freiburg, 2014.

 

Project: News Reception in Accordance with the Noise Trader Theory

 

Aims
  • Understand information processing of different investor types
  • Examine the effect of news sentiment on crude oil prices for different investor types in accordance with the Noise Trader Theory

 

Method
  • Data basis: Thomson Reuters News Archive for Machine Readable News 
  • Data utilization: Apply  sentiment analysis techniques to extract subjective information from news
  • Data analysis: Decompose the crude oil price with a Kalman filter into a fundamental price component and a noise residual, and regress a standardized news sentiment score on both decomposed oil price components

 

Results
  • News sentiment has a significant positive effect on the noise residual as suggested by the noise trader approach, but also on the fundamental price in contrast to the Noise Trader Theory

News reception with noise trader theory 

  • The effect of new sentiment is stronger on the noise residual than on the fundamental price

 

Alfano S, Feuerriegel S, Neumann D: Is News Sentiment More than Just Noise?, 23rd European Conference on Information Systems (ECIS 2015), Münster, Germany, May 26-29, 2015.

Alfano S, Feuerriegel S, Neumann D: Do Pessimists Move Asset Prices? Evidence from Applying Prospect Theory to News Sentiment, Working Paper, University of Freiburg 2015.
 

Project: Electricity Price Forecasting

 

Aims
  • Forecasting of electricity prices for both intraday and day-ahead auctions
  • Improving existing forecast models using real weather data as exogenous predictors
  • Variable selection of important predictors, i.e. weather stations

 

Method
  • Data basis: German electricity prices, weather data (wind & temperature) across 52 weather stations
  • Estimation of various forecasting models and comparison of their forecasting accuracy 
  • Methods from classical time series analysis (ARMAX, etc.) and machine learning (LASSO, Support Vector Regression, Random Forest, etc.)
  • Methods for variable selection in order to find weather stations with the highest influence

 

Results
  • Forecasting models with weather data as exogenous predictors are significantly more accurate than classical models without
  • Variable selection improves the forecasting accuracy and reduces the time complexity considerably

Electricity Price Forecasting (VarImp)

  • Daily rolling re-estimation of models is necessary for yield accurate forecasts

 

Ludwig N, Feuerriegel S, Neumann D: Putting Big Data Analytics to Work: Feature Selection for Forecasting Electricity Prices using the LASSO and Random Forests, Journal of Decision Systems, 2015: 24(1).

Feuerriegel S, Riedlinger S, Neumann D: Predictive Analytics for Electricity Prices using Feed-Ins from Renewables, 2014 (22nd European Conference on Information Systems (ECIS 2014), Tel Aviv, Israel, June 9-11, 2014, Complete Research Paper).
 

Project: Electricity Auction Design in Germany

Aims
  • Analysis of the EPEX SPOT electricity market in Germany; especially of the introduction of 15 minute contracts on 14/12/2011
  • Evaluation of the impact of 15 minute contracts on electricity price

 

Method
  • Data basis: electricity prices, wind volumes, load
  • Application of the Causal Impact method to proof causality and assess impact

 

Results
  • Introduction of 15 minute contracts has increased flexibility on SPOT market and reduced prices

Quick and dirty causal impact

  • Wind may be used as contemporaneous covariant for electricity prices

 

Project: Asymmetric Information Processing of Financial News Sentiment

 

Aims

  • Understanding processing of information with different polarity (negative vs. positive) by different investor types (extracted via Kalman filtering)

  • Finding evidence of asymmetric information processing and the negativity bias

 

Method

  • Data basis: Thomson Reuters News Archive for Machine Readable News

  • Utilizing sentiment analysis to extract subjective information from news

  • Applying quantile regressions to study the differential effect of positive and negative sentiment (negativity bias) on different investor types

 

Results
  • Different investor types process information differently

  • Informed investors weigh news with a very positive or negative sentiment less, uninformed investors weigh negative information more than positive confirming the negativity bias

index.jpg