Trading on News Intra-Day: Early Observations and Experiences Petter Kolm & Jun Wu Courant Institute (NYU) & the Heimdall Group, LLC kolm@cims.nyu.edu Results Are Preliminary Please Do Not Cite without Authors Permission Alpha Generation Using News Sentiment Data Deltix & RavenPack September 27, 2012 New York City TRADING ON NEWS INTRA-DAY: EARLY OBSERVATIONS AND EXPERIENCES, VER. 9/27/2012. P. KOLM 1
Presenter Biography Petter Kolm is the Director of the Mathematics in Finance Masters Program and Clinical Associate Professor at the Courant Institute of Mathematical Sciences, New York University and the Principal of the Heimdall Group, LLC. Previously, Petter worked in the Quantitative Strategies Group at Goldman Sachs Asset Management where his responsibilities included researching and developing new quantitative investment strategies for the group's hedge fund. Petter coauthored the books Financial Modeling of the Equity Market: From CAPM to Cointegration (Wiley, 2006), Trends in Quantitative Finance (CFA Research Institute, 2006), Robust Portfolio Management and Optimization (Wiley, 2007), and Quantitative Equity Investing: Techniques and Strategies (Wiley, 2010). He holds a Ph.D. in mathematics from Yale, an M.Phil. in applied mathematics from Royal Institute of Technology, and an M.S. in mathematics from ETH Zurich. Petter is a member of the editorial board of the Journal of Portfolio Management (JPM), International Journal of Portfolio Analysis and Management (IJPAM), Journal of Investment Strategies (JOIS), and the board of directors of the International Association of Financial Engineers (IAFE). As a consultant and expert witness, he has provided his services in areas such as algorithmic and quantitative trading strategies, econometrics, forecasting models, portfolio construction methodologies incorporating transaction costs, and risk management procedures. Email: Web: kolm@cims.nyu.edu http://www.theheimdallgroup.com http://www.cims.nyu.edu/~kolm TRADING ON NEWS INTRA-DAY: EARLY OBSERVATIONS AND EXPERIENCES, VER. 9/27/2012. P. KOLM 2
Outline! Introduction o Why this project o Our other projects in HFT: Our research agenda o Related literature! Trading signal & preliminary results! Conclusions & next steps The authors thank RavenPack for the news data and Deltix for the market data and the Deltix Product Suite (TimeBase, QuantOffice and QuantServer) used in this study TRADING ON NEWS INTRA-DAY: EARLY OBSERVATIONS AND EXPERIENCES, VER. 9/27/2012. P. KOLM 3
Introduction TRADING ON NEWS INTRA-DAY: EARLY OBSERVATIONS AND EXPERIENCES, VER. 9/27/2012. P. KOLM 4
Big Picture: Our HFT Research Agenda! Replicate common HFT strategies o Market making o Pairs trading and statistical arbitrage o Main research questions:! What are the strategy performance and risk characteristics?! How do these depend on parameters such as latency, rebates, trading venues?! How are different strategies correlated?! Back testing methodologies in HFT o Many challenges: rebates, market impact, timestamp issue, simulation of trade execution (market orders, limit orders)! Market impact models for HFT! TRADING ON NEWS INTRA-DAY: EARLY OBSERVATIONS AND EXPERIENCES, VER. 9/27/2012. P. KOLM 5
This Project (1/2): Overview! How we got into doing this project! Main questions that we are interested in: o Are news data & information useful for intraday trading? o Do (all) stocks react to (all) news releases? o How quickly do stock prices react to news releases? Some preliminary & partial answers:! Stocks show and quick reaction to news (often in the order of minutes or less)! Useful for intraday trading! Stocks react to news releases, but they do not seem to react (impact/magnitude) in the same way (of different types of news)! TRADING ON NEWS INTRA-DAY: EARLY OBSERVATIONS AND EXPERIENCES, VER. 9/27/2012. P. KOLM 6
This Project (2/2): Disclaimers! Research study is very preliminary! We have only been working with Ravenpack news data for a few weeks o There are many signals / modifications that we have not had a chance to look into o We are novices when it comes to news-based trading strategies! Data series we used are short (8 months) In short, there is a lot more work for us to do! TRADING ON NEWS INTRA-DAY: EARLY OBSERVATIONS AND EXPERIENCES, VER. 9/27/2012. P. KOLM 7
Some Related Literature Intraday trading:! Adams, McQueen et al. (1999) o Studied the relationship between unanticipated inflation (news) and size-based stock portfolios o Large & medium do respond in 15 minutes or less, Small cap stocks show weak or no response! Busse and Clifton Green (2002) 1 o Trades on the Morning Call and Midday Call segments from CNBC TV o Stock prices respond to reports within seconds of mention! Positive reports fully incorporated within 1 minute o Trading intensity doubles in the first minute 2! Mittermayer (2004) 3 o Categorizes news into three categories (1) good news, (2) no movers, (3) bad news using statistical learning/training o Trades 2 minutes after news release. Liquidates position after 58 minutes o Somewhat profitable to trade on good news; less so for shorting on bad news! Wang, Yang et al. (2006) o Looks at effects of monetary policy surprises on returns, volatilities, trading and volumes, and bid-ask spread of SPY and MDY o Overreaction of the SPY to unanticipated changes in the federal funds rate target in the first 5 minutes o 25 bps unanticipated cut in the federal funds rate! Increase in the SPY (by 1.2%) and MDY (by 1.6%), respectively TRADING ON NEWS INTRA-DAY: EARLY OBSERVATIONS AND EXPERIENCES, VER. 9/27/2012. P. KOLM 8
o 25 bps unanticipated decline (rise) in the 12M Eurodollar futures rate! Increase (or decrease) of SPY(0.71%) and MDY (0.40%), respectively o Study suggests that the market reacts more strongly to surprises in the future direction of monetary policy during the monetary tightening period and that the impact of monetary policy surprises depends on their sizes! Miao, Ramchander et al. (2011) o Examine the role of the 8:30am and 10:00am macroeconomic news announcements in explaining large and significant discontinuities in intraday futures prices on the Dow, Nasdaq and S&P 500 indices o Document strong relationship between the two sets of morning economic news releases and jumps in equity index futures prices o Good (bad) economic news is followed by positive (negative) jumps and that the responses are asymmetric; bad news has a larger impact on returns than good news! Füss, Mager et al. (2011) ( extension of Andersen, Bollerslev et al. (2002)) o Examine intraday price discovery processes of the German stock, bond, and U.S. dollar/euro FX markets, Jan. 2006 - Dec. 2008 o Analyze how quickly asset prices incorporate new information and react to macroeconomic news from the U.S. and Germany o Response patterns tend to be gradual, lasting from 1 to 8 minutes o New information affects market in the following order: FX! bonds! stocks TRADING ON NEWS INTRA-DAY: EARLY OBSERVATIONS AND EXPERIENCES, VER. 9/27/2012. P. KOLM 9
RavenPack:! Most studies conducted using daily data (close-to-close)! Hafez (2011): o Event study on analyst ratings, price target upgrade/downgrade, earnings, earnings guidance, dividends, credit ratings, corporate structure, legal issues, insider trading, product issues, M&A activity, buybacks, etc. o Market response approach (i.e. how has the market responded most recently to a given event category) as an adaptive refinement approach improves the risk-adjusted performance of the benchmark strategy in 76% of cases! Hafez (2012) o S&P 500, May 2005 - Dec..2009 o Company relevance and event novelty are important elements of a news-based strategy o Sentiment index delta constructed from trailing sentiment index [moving average of event sentiment score (ESS)] TRADING ON NEWS INTRA-DAY: EARLY OBSERVATIONS AND EXPERIENCES, VER. 9/27/2012. P. KOLM 10
Trading Signal & Preliminary Results TRADING ON NEWS INTRA-DAY: EARLY OBSERVATIONS AND EXPERIENCES, VER. 9/27/2012. P. KOLM 11
Data Market data (provided by Deltix):! Constituents 4 of the S&P 500 index! Time period: Jan. 2 Aug. 6, 2009 (150 trading days)! 1 minute-bar price data: open, close, high, low News data (provided by Ravenpack):! RavenPack tracks around 28,000 companies globally (more than 98% of the investable global market)! News classified into 278 event categories in 22 groups (such as equity-actions, M&As, product services, earnings, credit ratings)! News records are time-stamped to the millisecond and contain data for sentiment, novelty, relevance, event categories, and other news analytics! TRADING ON NEWS INTRA-DAY: EARLY OBSERVATIONS AND EXPERIENCES, VER. 9/27/2012. P. KOLM 12
Mean and Std. Dev. of the Number of News per Day (Relevance = 100) Total number of news over the period: 100760 Average number of news per day: 5 452 TRADING ON NEWS INTRA-DAY: EARLY OBSERVATIONS AND EXPERIENCES, VER. 9/27/2012. P. KOLM 13
Average Number of Good vs. Bad News per Day (Relevance = 100)! TRADING ON NEWS INTRA-DAY: EARLY OBSERVATIONS AND EXPERIENCES, VER. 9/27/2012. P. KOLM 14
Our Trading Signal Methodology (1/2) Event Sentiment Score (ESS)! Score between 0 and 100 o! 50 good o " 50 bad! Represents news sentiment for a given company based on various proxies sampled from the news! Determined by RavenPack by systematically matching stories typically rated by financial experts and analysts as having short-term positive or negative share price impact 6! Algorithm delivers a score for more than 330 types of news events (from product recalls to earnings announcements, and much more...)! TRADING ON NEWS INTRA-DAY: EARLY OBSERVATIONS AND EXPERIENCES, VER. 9/27/2012. P. KOLM 15
Our Trading Signal Methodology (2/2) Formation Period! Delay! Holding Period [ t - lookback, t ] [ t, t + delay ) [ t + delay, t + delay + hold ) Trading signal! RavenPack s Average Event Sentiment (AES) applied intraday o Calculate the ratio between the positive news count and the total news count over the formation period! Split stocks into 3 groups o Good sentiment (AES > 50) stocks are split into 2 equally sized portfolios o Bad sentiment ( AES " 50) stocks are split into 2 equally sized portfolios o No news over formation period! Rebalance at the end of the holding period! We vary lookback, delay and hold. No overlapping portfolios (at this point) TRADING ON NEWS INTRA-DAY: EARLY OBSERVATIONS AND EXPERIENCES, VER. 9/27/2012. P. KOLM 16
Preliminary Results Please refer to spreadsheets:! BucketRet_SS_2P.xlsx! BucketRet_SS_4P.xlsx! TRADING ON NEWS INTRA-DAY: EARLY OBSERVATIONS AND EXPERIENCES, VER. 9/27/2012. P. KOLM 17
Conclusions & Next Steps What we have done...! We have constructed an intraday trading signal based on the Event Sentiment Score (ESS) and Average Event Sentiment (AES) for each stock in the S&P 500 index! Intraday formation and holding periods delivers positive risk-adjusted returns What we are planning to do...! Trading signal - Improvements to be incorporated o Apply Event Novelty Score (ENS)! Score between 0 and 100 that represents how new or novel a new story is over a 24 hour time window 7 (see, Hafez (2012))! Realistic trading simulation o Real-time hedging of market and sectors o Position sizing based on risk/reward trade-off o Incorporate transaction costs (market impact) TRADING ON NEWS INTRA-DAY: EARLY OBSERVATIONS AND EXPERIENCES, VER. 9/27/2012. P. KOLM 18
! Appendix TRADING ON NEWS INTRA-DAY: EARLY OBSERVATIONS AND EXPERIENCES, VER. 9/27/2012. P. KOLM 19
When Do News Occur? (1/2) Important to understand when events occur:! Pre-market hours! Market hours! After-market hours! How the market responds or interprets company specific events depends on the prevailing market environment (market view on such events)! TRADING ON NEWS INTRA-DAY: EARLY OBSERVATIONS AND EXPERIENCES, VER. 9/27/2012. P. KOLM 20
When Do News to Occur? (2/2)! Varies from Group to Group Source: Hafez (2011) TRADING ON NEWS INTRA-DAY: EARLY OBSERVATIONS AND EXPERIENCES, VER. 9/27/2012. P. KOLM 21
References Adams, G. L., G. McQueen and R. Wood (1999). "The Effects of Inflation News on High Frequency Stock Returns." Available at SSRN 176268. Andersen, T. G., T. Bollerslev, F. X. Diebold and C. Vega (2002). Micro Effects of Macro Announcements: Real-Time Price Discovery in Foreign Exchange, National Bureau of Economic Research. Busse, J. A. and T. Clifton Green (2002). "Market Efficiency in Real Time." Journal of Financial Economics 65(3): 415-437. Füss, R., F. Mager and L. Zhao (2011). "Price Discovery and Information Transmission Processes among Asset Markets: An Ultrahigh-Frequency Perspective." Available at SSRN 1966958. Hafez, P. A. (2011) "Event Trading Using Market Response." Hafez, P. A. (2012). "How News Events Impact Market Sentiment." The Handbook of News Analytics in Finance: 129-146. Miao, H., S. Ramchander and J. Zumwalt (2011). "Information Driven Price Jumps and Trading Strategy: Evidence from Stock Index Futures." Available at SSRN 1927966. Mittermayer, M. A. (2004). Forecasting Intraday Stock Price Trends with Text Mining Techniques, IEEE. Wang, T., J. Yang and J. Wu (2006). "Central Bank Communications and Equity Etfs." Journal of Futures Markets 26(10): 959-995. TRADING ON NEWS INTRA-DAY: EARLY OBSERVATIONS AND EXPERIENCES, VER. 9/27/2012. P. KOLM 22
Endnotes!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 1 The Morning Call aired between 11:05 a.m. and 11:10 a.m. (EST) and the Midday Call aired between 2:53 p.m. and 2:58 p.m. (EST). Their data set consists of 322 stock reports over 84 different trading days from the period June 12 - Oct. 27, 2000. Traders who execute within 15 seconds of the initial mention make small but significant profits by trading on positive reports during the Midday Call. 2 They find a significant increase in buyer- (seller-) initiated trades after positive (negative) reports. 3 They use TAQ data from Jan. 1, 2002 Dec. 31, 2002 and press releases published by PRNewswire during that same time period. 4 We removed the following 21 companies from the S&P that in the dataset we are provide contained missing data: ABC (Amerisourcebergen Corp), AIG (American International Group Inc), TWC (Time Warner Cable Inc), TWX (Time Warner Inc), DNR (Denbury Resources Inc), DO (Diamond Offshore Drilling Inc), DV (Devry Inc Del), FTI (F M C Technologies Inc), HRL (Hormel Foods Corp), NU (Northeast Utilities), ORLY (O Reilly Automotive Inc), PCS (Metropcs Communications Inc), PWR (Quanta Services Inc), RHT (Red Hat Inc), SNI (Scripps Networks Interactive Inc), TDC (Teradata Corp De), THC (Tenet Healthcare Corp), VTR (Ventas Inc), WDC (Western Digital Corp), WM (Waste Management Inc Del), and WPO (Washington Post Co). 5 Calculated using a total of 223 days as there can be news on weekends as well. 6 The strength of the score is derived from training sets where financial experts classified company-specific events and agreed these events convey positive or negative sentiment and to what degree. Their ratings are encapsulated in an algorithm that generates a score range between 0-100 where higher values indicate more positive sentiment while lower values below 50 show negative sentiment. 7 The first story disclosing an event about a company is considered to be the most novel and receives a score of 100. Subsequent stories about the companies event receive lower scores following a decay function (100, 75, 56, xxxxx). Stories outside the 24 hour window but similar to a story in a chain of events receive a score of 0.! TRADING ON NEWS INTRA-DAY: EARLY OBSERVATIONS AND EXPERIENCES, VER. 9/27/2012. P. KOLM 23