Forex-foreteller: A News Based Currency Predictor Fang Jin, Nathan Self, Parang Saraf, Patrick Butler, Wei Wang, Naren Ramakrishnan Department of Computer Science Virginia Tech Aug 13, 2013
2 EMBERS Funded by Intelligent Advanced Research Projects Activity (IARPA) Primarily Interested in making predictions about Latin American Countries The primary prediction areas are as follows: Civil Unrest Events Influenza Like Illness Events Rare Diseases Events Elections Financial Events
3 Foreign Exchange Market Most liquid financial market in the world Average daily turnover was USD 3.98 trillion in April 2010 Growth of approximately 20% as compared to 2007 United States GDP is around USD trillion Operates 24 hours a day except on weekends Geographically Dispersed Traders include large banks, central banks, institutional investors, currency speculators, corporations, governments and retail investors A variety of factors effect exchange rate: Economic Factors Political Conditions Market Psychology
4 Related Work Fundamental Analysis Analyses economic health of a country Employment Reports Inflation Productivity Trade Growth Technical Analysis Mathematical Techniques like VAR, ARCH, GARCH etc Based on Past Trends of financial indicators Can’t rely on just one type. Have to use a combination of both the techniques
5 Our Approach Bloomberg News Interest Rates Inflation Unanticipated News Past Currency Values Past Stock Values Linear Regression Model Final Prediction FundamentalTechnical
6 System Framework
7 Language Modeling Different Types of News Latent Dirichlet Allocation Model to identify different topics Top 30 topics are Identified Out of 30 topics, manually identify topics of Interest List of Interesting topics
8 Topic Clustering Identify trending topics by tracking topic distribution movement over time
9 Sentiment Analysis Interest Rate Increase/Decrease Inflation Increase/Decrease Unanticipated News
10 Linear Regression Interest Rates Inflation Unanticipated News Past Currency Values Past Stock Values Linear Regression Model Final Prediction Where: Δ c is currency change Δ r is interest rate change Δ f is interest rate change Δ s is currency change Δ e is currency change β r, β f, β s, β e are respective weights
11 Off-line Components
12 Online Components Displays the generated alerts and associated Audit trails for user analysis
13 EMBERS Visualizer Link:
14 Other EMBERS Products Civil Unrest PredictorInfluenza Like Illness Predictor Rare Diseases Predictor Ablation Visualizer Link:
15 Fang Jin: Parang Saraf: