Reinforcement in International Financial Markets Ken Froot, HBS Xiaoxia Lou, University of Delaware Gideon Ozik, EDHEC Business School Ronnie Sadka, Boston College Siyi Shen, Boston College March 2018 1
Research Question An avenue through which information is gathered, processed, and disseminated Large amount of data are generated by media daily Academic research has focused on direct effects coverage can predict returns Mainly focused on individual stocks and US aggregate equity This work studies the interaction of media and asset prices Individual stocks Aggregate equity markets 2 Currencies
How Does the Interact with Asset Prices? How to measure optimism / pessimism? Asset prices is one indicator Look at media sentiment! This work advances a simple concept: When return and sentiment reinforce one another There is unusually high optimism, which results with overreaction Expected reversal w/o media Abnormal return Expected reversal with media 3
Related Literature The role and content of media and its impact on asset prices: 4 e.g., Tetlock (2007), Tetlock, Saar-Tsechansky, and Macskassy (2008), and Chen, De, Hu, and Hwang (2014) Short-term return autocorrelation: e.g, Jegadeesh (1990), Lehman (1990), Jegadeesh and Titman (1995), Copper (1999), and Avramov, Chordia, and Goyal (2006) Information dissemination in financial market: e.g., Chan (2003), Tetlock (2010), and Griffin, Hirschey, and Kelly (2011) Investor behavioral biases: e.g., Daniel, Hirshleifer, and Subrahmanyam (1998), Barber and Odean (2008), and Solomon, Soltes, and Sosyura (2014)
The Power of the Wealth of information A careful examination of the data and the correction for various effects General Specialized Corporate Communications Social What is the world saying? What is the industry saying? What are companies saying? What are people saying? 5
The Data Countries FX and Equity indices 12 developed market currency: AUD, CAD, CHF, DKK, EUR, GBP, ILS, JPY, NOK, NZD, SEK, SGD +2 developed market equity: HKD, USD 17 emerging markets: ARS, BRL, CLP, CNY, COP, EGP, IDR, INR, MXN, MYR, NGN, PHP, PLN, RUB, THB, TRY, ZAR Coverage Currencies Equity indices Other Asset Classes Large-cap stocks Commodities Sentiment Scoring Textual analysis 6
The Data Cont d Number of articles covering firms Number of articles by country of source (FX, Country) count Zoom 100k 50k 0k Non SP5 0 0 NonUS SP5 0 0 Jan '16 Jul '16 Jan '17 Jul '17 USA Euro Zone UK India Japan China Canada Hong Kong Australia Mexico South Africa Malaysia Thailand Brazil New Zealand Switzerland Singapore Russia - 30,000 60,000 90,000 120,000 150,000 FX / Country articles by year FX / Country articles by source type 7
Brexit Vote (June 23, 2016) 1-day abnormal country equity sentiment Prior to Brexit vote, sentiment seemed mostly positive Once Leave was announced, global sentiment turned sharply negative, with UK, European countries, the Americas and Australia leading the way In contrast, Russia and China exhibit a positive sentiment shock 8
US Presidential Election (Nov 8, 2016) 1-day abnormal FX sentiment The extent of the results became clear only after midnight ET. Therefore, media on 11/8/2016 does not reflects the surprising results whereas media coverage on 11/9/2016 reflect the full extent of the results While world sentiment turned negative overall, a few countries displayed positive sentiment, notably, Russia and Turkey 9
French Presidential Election (April 23, 2017) 1-day abnormal country equity sentiment The results of the first round indicated strong performance of Emmanuel Macron, the centerleaning candidate, alleviating concerns of anti-european pressures Other than a few exceptions (e.g., Portugal, Poland), country equity sentiment reacted positively 10
Tests using Portfolio s First examine the relative autocorrelation in the different markets Form 10-day-ladder portfolios based on past weekly returns Then, add past weekly media sentiment Sample: March 2013 April 2017 11
Reinforcement Portfolio Sorts 10-day-ladder portfolios sorted by past weekly return and sentiment s s Sentiment Sentiment Reversal Reinforcement FX Developed 1 Country Equity 2 Large Stocks 3 +2.04% [1.94] +0.83% [0.91] -2.90% [-1.91] -3.76% [-2.10] -1.15% [-1.33] -1.72% [-1.92] +2.12% [1.78] +0.31% [0.27] -2.53% [-1.56] -5.11% [-2.49] 0.57% [0.44] -3.01% [-2.57] +2.25% [2.01] +1.09% [1.00] -3.34% [-1.75] -4.57% [-2.09] -1.02% [-1.03] -2.32% [-2.07] 1 Sentiment measured from FX media; 2 Sentiment measured from FX media; 3 Sentiment measured from stock equity media 12
Portfolios in event time: Stocks Reinforcement effect return and high sentiment leads to low return return and low sentiment leads to high return 0.00% 0.09% 2.50% 0.01% -0.50% -1.00% low sentiment 0.07% 0.05% 2.00% 1.50% low sentiment -0.01% -0.03% -1.50% -2.00% high sentiment 0.03% 0.01% 1.00% 0.50% high sentiment -0.05% -0.07% -2.50% -5-4 -3-2 -1 0 1 2 3 4 5 6 7 8 9 10 Days -0.01% 0.00% -5-4 -3-2 -1 0 1 2 3 4 5 6 7 8 9 10 Days -0.09% 13
Reinforcement or Feedback? Empirical Design Decomposition of Expected and Unexpected return and sentiment:,,,,,,,,,,,, In-sample estimation, per asset Expected components explain a small fraction of total variance: Average R 2 ranges between 1.23% to 2.37% 15
Portfolio Sorts Expected Components 10-day-ladder portfolios sorted by past weekly expected return and sentiment s s Sentiment Sentiment Reversal Reinforcement FX Developed Country Equity Large Stocks -2.06% [-2.70] -0.24% [-0.27] 2.09% [1.89] 3.06% [2.38] 1.29% [1.51] 1.01% [1.31] -1.99% [-1.67] -2.45% [-2.30] 4.34% [2.61] 2.47% [1.20] 3.96% [2.78] 0.48% [0.43] -4.84% [-3.56] -5.16% [-5.80] 9.99% [6.20] 9.76% [4.52] 5.07% [4.93] 4.93% [5.50] The expected components generate continuation; no reinforcement effect 16
Portfolio Sorts Unexpected Components 10-day-ladder portfolios sorted by past weekly unexpected return and sentiment s s Sentiment Sentiment Reversal Reinforcement FX Developed Country Equity Large Stocks 2.03% [1.92] 0.69% [0.72] -2.76% [-1.97] -4.37% [-2.55] -0.38% [-0.39] -2.34% [-2.79] 3.14% [2.64] 0.75% [0.64] -3.99% [-2.51] -6.69% [-3.35] -0.36% [-0.29] -3.54% [-3.25] 2.47% [2.30] 1.89% [1.63] -4.35% [-2.23] -4.70% [-2.23] -2.13% [-1.96] -2.23% [-2.02] The unexpected components generate reversal; The results are consistent with reinforcement rather than feedback 17
Macroeconomic News and Earnings Announcements Construct portfolios excluding news dates in formation period s s Sentiment Sentiment Reversal Reinforcement FX Developed Country Equity Large Stocks +1.92% [2.01] +019% [0.19] -2.17% [-1.57] -3.40% [-2.04] -0.64% [-0.74] -1.48% [-1.66] +2.34% [1.58] +0.93% [0.67] -3.26% [-1.70] -5.40% [-2.13] -0.21% [-0.14] -3.06% [-2.18] +2.63% [2.39] +1.00% [0.94] -3.63% [-1.88] -5.39% [-2.48] -0.87% [-0.84] -2.76% [-2.45] The results are not due to main information events 18
Additional Analyses Including additional sources for FX and Country equity Alternative measures of sentiment Cross-sectional regressions using quartile dummies Different types of media Strong in local media Effect is stronger for large caps, highly covered by the media Calculation of risk-adjusted returns (per asset class) Emerging markets and Commodities 19
Relation to Intensity of Coverage Sentiment + Sentiment + 0.00% 0.12% 0.01% 0.02% 0.03% Coverage 0.10% 0.08% Coverage 0.04% 0.06% 0.05% 0.06% 0.07% Coverage 0.04% 0.02% Coverage 0.08% 0 1 2 3 4 5 6 7 8 9 10 Day 0.00% 0 1 2 3 4 5 6 7 8 9 10 Day er media coverage intensifies reinforcement 20
Relation to Liquidity: Individual Stocks Sentiment + Sentiment + 0.02% 0.00% Small 0.10% 0.08% -0.02% 0.06% Large -0.04% 0.04% -0.06% -0.08% Large 0.02% 0.00% Small -0.10% 0 1 2 3 4 5 6 7 8 9 10 Day -0.02% 0 1 2 3 4 5 6 7 8 9 10 Day Reinforcement effect more prominent in large caps Liquid firms attract more investors 21
What happens When You Combine All of These? Strategy Construct the 2 x 2 return/sentiment portfolios Weekly formation period, skip one day, 10-day ladder Long: low return low sentiment Short: high return high sentiment Start day: 3/1/2013, End day: 4/27/2017 Performance Examine FX (developed), Country equity (all), and firm equity (large caps) separately Combine all (volatility weighted) 1.30 1.20 1.10 1.00 0.90 Combined; IR = 1.75 22
Summary and Conclusion is a fruitful avenue for research Reinforcement effect in financial markets Robust results 23