Purging Investor Sentiment Index from Too Much Fundamental Information

Transcription

1 Purging Investor Sentiment Index from Too Much Fundamental Information Liya Chu Qianqian Du Jun Tu Singapore Management University (Chu, Tu) Southwestern University of Finance and Economics (Du) Lingnan (University) College Sun Yat-sun University November 7th, 2017 Presented by Jun Tu

2 Sentiment matters! Sentiment: excessively optimistic or pessimistic not justified by the facts at hand [Baker and Wurgler (2006)] Long history: Keynes (1936) Sentiment => asset price away from its fundamental value due to limits of arbitrage (e.g., investor myopia (mutual fund manager agency problem), short-sell constraint, margin constraint) Noise trader model: De Long et al. (1990, JPE); Barberis et al. (1998, JFE); Hong and Stein (1999, JF), etc. Recent theoretical models: Barberis, Greenwood, Jin and Shleifer (JFE; 2015); Benhabib, Liu, and Wang (JFE; 2016) Anecdotal observations of sentiment-driven bubbles and crashes 1961 crash of growth stocks; electronics bubble early 1980s biotech bubble; subprime crisis late 1990s Internet bubble Jun TU Purging Investor Sentiment Index November 7th, / 31

3 Sentiment unobservable though! Construct a proxy? Sentiment is not directly observable A popular proxy: Baker and Wurgler (2006, JF) investor sentiment index The first principal component (PC) of the 6 sentiment proxies Closed-end fund discount rate, CEFD Share turnover, TURN Number of IPOs, NIPO First-day returns of IPOs, RIPO Dividend premium, PDND Equity share in new issues, EQTI Empirically, explains well the cross-sectional stock returns Consistent to economic intuition, BW sentiment index strongly predicts stocks speculative, hard to arbitrage, or in the short legs of long-short strategies Jun TU Purging Investor Sentiment Index November 7th, / 31

4 Is BW index influential? Yes. Is it clean? No!! Influential: 1980 google citations! and rising every day! e.g., Baker and Wurgler (2006, 2007), Baker, Wurgler, and Yuan (2012, JFE), Stambaugh, Yu, and Yuan (2012, JFE) six components in BW index correlated with economic environment. on top of sentiment, number of IPOs economic condition (Ritter and Welch, 2002) Baker and Wurgler smart, notice this, and removed a few business cycle variables. However, orthogonal BW index has a 0.97 correlation with original BW index. Sibly et al. (2015). Jun TU Purging Investor Sentiment Index November 7th, / 31

5 Investor Sentiment and Economic variables Regression of BW index on fundamental variables coef. OLS t-stat NW t-stat fundamental variables Intercept Output and income Employment Housing Consumption, orders and inventories Money and credit Exchange rates Inflation Consumption wealth ratio GDP growth m Treasury Bill Default spread Term spread Dividend yield Zero return ratio R-square 62.51% adj R-square 61.60% Baker and Wurgler s investor sentiment - 60% can be explained by fundamental variables Jun TU Purging Investor Sentiment Index November 7th, / 31

6 BW index driven by fundamental => Literature misleading; Purged index needed! Do we care BW clean or not? Of course! If BW index is dominated by fundamental information and a proxy for systematic risks, then numerous papers adopting BW sentiment index as a behavioral variable could all be misleading!! Baker et al.(2012, JFE) - international; Stambaugh et al.(2012, JFE) - anomalies; Hribar et al.(2012, MS) - analysts forecast; Antoniou et al.(2015, MS) - slope of security market line; etc. Baker and Wurgler (2006; 2007) issue an warning on BW index contaminated by fundamentals! Jun TU Purging Investor Sentiment Index November 7th, / 31

7 Purged investor sentiment: clean and better! Orthogonal Baker and Wurgler (2006) sentiment index to a wide range of economic fundamentals For each of the 6 sentiment proxies, we remove 14 fundamental factors, including seven first principal components from 109 macroeconomic variables, consumption-to-wealth ratio, GDP growth, the yield on three-month Treasury Bills, the default spread, the term spread, dividend yield and liquidity risk factor. Econometrically, using a Patial Least Square (PLS) method to eliminate common noises of the proxies PLS [6 purged sentiment proxies] Purged Sentiment Index [IS-P] IS-P has similar or stronger forecasting power than Baker and Wurgler (2006) sentiment index after removing 60% (fundamental) information It is comparable to or better than the original BW index, and outperforms other survey-based sentiment indices, such as consumer confidence index, Gullep Survey index; Jun TU Purging Investor Sentiment Index November 7th, / 31

8 Latent Factor Structure of Sentiment Proxies Each individual sentiment proxy has a factor structure x i,t = η i,0 + η i,1 S t + η i,2 E t + e i,t, for i = 1,..., N, (1) where x i,t : Individual sentiment proxy; S t : Unobservable investor sentiment; E t : Common approximation error component or Common noise Baker and Wurgler (2006) S t is estimated as the first principle component (PC) of the x i,t s PC explains the largest fraction of the total variations in x i,t s PC can NOT separate S t from E t the larger the variance the E t, the more important role will it play in the PC can remove e i,t part but not the common error E t part Jun TU Purging Investor Sentiment Index November 7th, / 31

9 Partial Least Squares (PLS) CAN Remove Common Error Realized stock return then is R t+1 = α + βs t + ε t+1, (2) where S t is the true but unobservable investor sentiment, ε t+1 : unpredictable shock and unrelated to S t This paper applies the PLS method for extracting S t in (1) S t is estimated from six sentiment proxies x i,t s according to covariance with future stock returns Intuition: utilizing expected stock return to proxy for purged sentiment S t exploiting the factor nature of the joint system (2) and (1) can effectively discard irrelevant components like E t and e i,t Wold (1966, 1975; Chemistry) and Kelly and Pruitt (2012, 2013) Jun TU Purging Investor Sentiment Index November 7th, / 31

10 PLS produces Purged Sentiment with Common Error Removed Two-step PLS estimation strategy for linear system (1) and (2) For each period t, run the cross-sectional regression x i,t = c t + IS PLS t ˆπ i + v i,t, for t = 1,..., T, (3) ISt PLS, the regression coefficient in (3), is the purged investor sentiment ˆπ i in (3) is estimated via the time-series regressions x i,t 1 = π i,0 + π i R t + u i,t 1, for i = 1,..., N, (4) where π i captures the sensitivity of x i,t 1 to S t 1 proxied by R t Jun TU Purging Investor Sentiment Index November 7th, / 31

11 Relative weight of TURN in sentiment measures The loadings of the six purified sentiment indicators in IS-P: IS-P t = 0.17 CEFDres t TRUNres t NIPOres t RIPOres t PDNDres t EQTIres t (5) The loadings of the six raw sentiment indicators in BW index: BW t = 0.28 CEFD t TRUN t NIPO t RIPO t PDND t EQTI t (6) IS-P allocates much smaller relative loading on TURN than the BW index does: Although TURN is among the three most important indicators out of the six sentiment indicators in the BW index, it becomes almost the least important indicator in the IS-P index The loading of the TURN indicator in IS-P seems declining over the recent years (based on a rolling window approach). However, the loading of the TURN indicator is still sizable. Jun TU Purging Investor Sentiment Index November 7th, / 31

12 Testing Portfolios Prior research: investor sentiment affects cross-section stock return, especially for stocks that are difficult to arbitrage or value. Following Baker and Wurgler (2006), we construct spread portfolios (i.e., high, medium and low) according to NYSE breakpoint of firm characteristics: size, age, dividend payment, earnings, tangible assets, R&D, sigma, external finance, sales growth, and book-to-market ratio. Jun TU Purging Investor Sentiment Index November 7th, / 31

13 BW Investor Sentiment can predict cross section stock returns; BW 2006 paper claim that this is consistent with BW index capturing sentiment impact on Jun TU Purging Investor Sentiment Index November 7th, / 31 Is BW Sentiment Index working? Yes! 11 out of 16 is significant Long-Short No Control FF(t) BW BW coef p value coef p value ME High-Low Age High-Low Sigma Low-High E/BE > 0 < D/BE > 0 = PPE/A High-Low RD/A Low-High BE/ME High-Low EF/A Low-High GS Low-High BE/ME Med-Low EF/A Med-High GS Med-High BE/ME Med-High EF/A Med-Low GS Med-Low Combination

14 Is BW Sentiment Index still working after fundamental removed? No? 1 out of 16 is significant Long-Short No Control FF(t) BW BW coef p value coef p value ME High-Low Age High-Low Sigma Low-High E/BE > 0 < D/BE > 0 = PPE/A High-Low RD/A Low-High BE/ME High-Low EF/A Low-High GS Low-High BE/ME Med-Low EF/A Med-High GS Med-High BE/ME Med-High EF/A Med-Low GS Med-Low Combination NO: BW (removing fundamental) NO longer predicts cross section stock returns!? BW is fundamental, Not sentiment index? Numerous studies using behaviorial explanation Misleading?! Jun TU Purging Investor Sentiment Index November 7th, / 31

15 Again, is BW Sentiment Index still working after fundamental removed? YES! Why? 13 out of 16 is significant; noise is removed by using PLS method! Long-Short No Control FF(t) IS-P IS-P coef p value coef p value ME High-Low Age High-Low Sigma Low-High E/BE > 0 < D/BE > 0 = PPE/A High-Low RD/A Low-High BE/ME High-Low EF/A Low-High GS Low-High BE/ME Med-Low EF/A Med-High GS Med-High BE/ME Med-High EF/A Med-Low GS Med-Low Combination Jun TU Purging Investor Sentiment Index November 7th, / 31

16 Comparison BW is the regression residual of BW index on 6 macro variables; BW is the regression residual of BW index on 14 fundamental factors BW BW BW IS-P long-short no FF bootstrap p FF(t) bootstrap p long leg no FF bootstrap p FF(t) bootstrap p short leg no FF bootstrap p FF(t) bootstrap p Sentiment strongly forecasts cross section stock returns if fundamental is purged properly. Jun TU Purging Investor Sentiment Index November 7th, / 31

17 Figure 1 BW original index and the purged sentiment index IS-P The black solid line represents the purged sentiment index IS-P; the blue dashed line represents the original BW index Jun TU Purging Investor Sentiment Index November 7th, / 31

18 Macroeconomic Activities IS-P (BW) and macroeconomic activities Panel A: IS-P coef dependent variable (1) (2) (3) (4) Output and income [-0.15] Employment 0.15 [1.17] Housing [-0.29] Consumption, orders and inventories [-1.48] R2(%) Panel B: BW coef dependent variable (1) (2) (3) (4) Output and income [-1.73] Employment [-2.38] Housing 0.29 [3.28] Consumption, orders and inventories [-3.73] R2(%) Purged sentiment contains much less fundamental information compared with original BW index. Jun TU Purging Investor Sentiment Index November 7th, / 31

19 Robustness Check Firstly, we extract seven common factors from more than one hundred of macroeconomic variables using asymptotic principal component analysis. Secondly, we directly remove 130 variables [7 groups of macros in FRED-MD and 5 fundamental related variables]. We find that purged sentiment index are consistently successful in predicting cross-sectional stock return, while residual of BW index or PCA way constructed index fail to. Jun TU Purging Investor Sentiment Index November 7th, / 31

20 Further analysis: mutual fund flow IS-P and fund flow Panel A. purged sentiment (1) (2) IS-P(t) 0.23 [2.89] IS-P(t-1) 0.23 [2.79] R2(%) Panel B. BW sentiment (3) (4) BW(t) [-0.37] BW(t-1) [-0.74] R2(%) IS-P significantly predicts mutual fund flow while BW sentiment index fails. Jun TU Purging Investor Sentiment Index November 7th, / 31

21 Further analysis: survey-based sentiment measures IS-P and other survey based sentiment measures We obtain anxious index (AI) from Federal Reserve Bank of Philadelphia that measures the probability of a decline in real GDP, consumer sentiment index (ICS) from Michigan University, individual investor sentiment index (AAII) from American Association of Individual Investor survey and rescaled Gallup investor index GA from Galllup survey. We orthogonalize each survey based sentiment index to the 14 macroeconomic. After decomposing anxious index (AI), consumer sentiment index (ICS), individual investor sentiment index (AAII), and rescaled Gallup investor index GA, we derive AIres, ICSres, AAIIres and GAres respectively. Jun TU Purging Investor Sentiment Index November 7th, / 31

22 Further analysis: survey-based sentiment measures IS-P and other survey based sentiment measures (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Panel B. L-S combine AIres [-0.58] [-0.91] ICSres [0.31] [0.55] AAIIres [-1.63] [-1.38] GAres [-0.01] [-0.18] IS-P(AI) [3.68] [3.73] IS-P(ICS) [4.13] [4.11] IS-P(AAII) [3.16] [3.06] IS-P(GA) [3.31] [3.32] The predictive power of survey-based sentiment measures is much weaker than that of IS-P. Jun TU Purging Investor Sentiment Index November 7th, / 31

23 Validation Test IS-P and earnings announcement returns Panel A: IS-P Panel B: BW Decile Decile L M H L M H ME coef ME coef t-stat t-stat Age coef Age coef t-stat t-stat Sigma coef Sigma coef t-stat t-stat PPE/A coef PPE/A coef t-stat t-stat RD/A coef RD/A coef t-stat t-stat BE/ME coef BE/ME coef t-stat t-stat EF/A coef EF/A coef t-stat t-stat GS coef GS coef t-stat t-stat <= 0 > 0 <= 0 > 0 E/BE coef E/BE coef t-stat t-stat D/BE coef D/BE coef t-stat t-stat Earnings announcement returns are inversely related to purged sentiment index for speculative stocks (comparable to original BW index) Jun TU Purging Investor Sentiment Index November 7th, / 31

24 Validation Test IS-P and non-fundamental Q (e.g., Rhodes-Kropf et al. (2005), Grundy and Li (2010)) Panel C. IS-P mq dependent variable beta t-stat beta t-stat R 2 (%) L-S combine L combine S combine IS-P presents better predictability for the portfolio returns than mispricing component in Tobin s Q. Jun TU Purging Investor Sentiment Index November 7th, / 31

25 Economic Explanation Cash flow or discounst rate channel This table reports estimation results for the bivariate predictive regressions y t+1 = α + βis-p t + ϕdp t + ω t+1, y = d t+1 or dp t+1 where dp t+1 is the annual log dividend-price ratio for each short leg of characteristics based portfolios, d t+1 is the annual log dividend-growth rate for each short leg of characteristics based portfolios from July of year t to June of year t+1 (in percentage), IS-P t is the purged investor sentiment index in June of year t. d t+1 and dp t+1 are constructed following Cochrane (2008, 2011). We report the regression slopes, Newey-West t-statistics, as well as R 2 s. The sample period is from 1965 to y t+1 β t-stat ϕ t-stat R 2 (%) Panel A. combination portfolio dp d(%) Cash flow channel is the source for predictability. Jun TU Purging Investor Sentiment Index November 7th, / 31

26 Further Discussion on BW(5) Recently, the average NYSE share turnover (TURN) has been dropped as one of the six sentiment indicators at the Jeffrey Wurglers website, given that it seems TURN has become a bad sentiment indicator in the recent years. BW(5) and BW (5) are still highly correlated with BW index (with correlations of 0.97 and 0.94 respectively). BW (5) seems not purging the fundamental information content much given that it has a high correlation of 0.97 with BW(5) Jun TU Purging Investor Sentiment Index November 7th, / 31

27 Contamination of BW(5) index Regression of BW(5) index on fundamental variables Coef ols t-stat NW t-stat fundamental variables Intercept Output and income Labour market Housing Consumption,orders and inventories Money and credit Exchange rates Inflation Consumption wealth ratio GDP growth m Treasury Bill Default spread Term spread Dividend yield Zero return ratio R2(%) adj R2(%) Similar to the case of BW index, the BW(5) index is significantly contaminated by economic fundamentals as well. Jun TU Purging Investor Sentiment Index November 7th, / 31

28 Is BW(5) Sentiment Index still working after fundamental removed? NO: BW(5) (fundamentals removed) performs much worse than IS-P in forecasting cross-section returns. long-short long leg short leg BW(5) BW(5) BW(5) no FF 14 variables alternative 14 variables variables FF(t) 14 variables alternative 14 variables variables Jun TU Purging Investor Sentiment Index November 7th, / 31

29 The Impact of Sentiment on Anomalies - BW (5) Anomaly Long leg Short leg Long-short Panel A: High Low High High Low High High Low High BW (5) sentiment sentiment - low sentiment sentiment - low sentiment sentiment - low O-score Distress Net stock issues Composite equity issues Total accruals Net operating assets Momentum Gross profitability Asset growth Return on assets Investment-to-assets Combination Jun TU Purging Investor Sentiment Index November 7th, / 31

30 The Impact of Sentiment on Anomalies - IS-P Anomaly Long leg Short leg Long-short Panel B: High Low High High Low High High Low High IS-P sentiment sentiment - low sentiment sentiment - low sentiment sentiment - low O-score Distress Net stock issues Composite equity issues Total accruals Net operating assets Momentum Gross profitability Asset growth Return on assets Investment-to-assets Combination Jun TU Purging Investor Sentiment Index November 7th, / 31

31 Conclusion Baker and Wurgler (2006) sentiment index contains too much fundamental! A fundamental index? Propose a Purged BW index can forecasts cross section returns; And outperforms other sentiment predictors (BW original, BW new TURN dropped index, consumer confidence index, Gullep Survey, etc) Now forecasting power of Purged sentiment index is from a pure sentiment or non-fundamental forces Therefore, our study not only provide crucially needed support for the literature using the original Baker and Wurgler (2006) sentiment index as a behavioral indicator but also a pugged one should be used as a safer choice/higher bar against data mining! Jun TU Purging Investor Sentiment Index November 7th, / 31