400 likes | 532 Views
Information Risk and Momentum Anomalies. Chuan-Yang Hwang Nanyang Technological University, Singapore and Xiaolin Qian University of Macao. Motivation. Hwang and Qian (2011) construct a measure of information risk/asymmetry (ECIN) based on the price discovery of large trades.
E N D
Information Risk and Momentum Anomalies Chuan-Yang Hwang Nanyang Technological University, Singapore and Xiaolin Qian University of Macao
Motivation • Hwang and Qian (2011) construct a measure of information risk/asymmetry (ECIN) based on the price discovery of large trades. • ECIN has stronger power in predicting stock returns than all the well-known variables. • Failure to take into account of the information risk can lead to false discovery of anomalies. • In this paper, we illustrate this point with momentum anomalies.
Momentum Literature • Price Momentum • Jegadeesh and Titman (1993) document the phenomenon in which past winenrs continue to be winners, and past losers continue to be losers for up to 12 months. • Post Earnings Announcement Drift (PEAD • High standardized-unexpected-earnings (SUE) firms earn higher risk-adjusted returns than low SUE firms for up to 12 months. Ball and Brown (1968), Bernard and Thomas (1990). Mendenhall (2004), Chordia and Shivakumar(2006), Zhang (2008).
Momentum Literature • Price Momentum • Jegadeesh and Titman (1993) document the phenomenon in which past continue to be winners, and past losers continue to be losers for up to 12 months. • Post Earnings Announcement Drift (PEAD) • High standardized-unexpected-earnings (SUE) firms earn higher risk-adjusted returns than low SUE firms for up to 12 months. Ball and Brown (1968), Bernard and Thomas (1990). Mendenhall (2004), Chordia and Shivakumar(2006), Zhang (2008).
Momentum Literature • Rational Asset Pricing Models • Fama (1998): One of the biggest challenges to rational asset pricing. • Sadka (2006) Liquidity factor explains 60%-80% of the momentum in NYSE stocks. • Behavioural Models • Daniel, Hirshleifer and Subrahmanyam (1998), Barberis, Shleifer and Vishny (1998), Hong and Stein (1999). • Zhang (2006) Momentum is stronger among firms with larger information uncertainty.
Information Risk Hypotheses of Momentum • Losers (bad news firms) are less attractive to informed traders because it is more difficult and riskier to do arbitrage on them. • Losers (bad new firms) have lower information risk than Winners (good new firms) do. • If information risk is priced, then winner have higher returns than losers--known as momentum.
Testable Implications (I) • H(1) Losers have lower information risk than Winners do. • H(2) After adjusting for the information risk differential between Winners and Losers, momentum anomalies should weaken or disappear altogether.
Testable Implications (II) • High idiosyncratic volatility (IVOL) makes arbitrage on both Winner and Losers riskier, but the impact will be larger on Losers. • H(3) The information risk decreases with IVOL for both Winners and Losers. • H(4) The information risk differential between Winners and Losers, and hence the strength of momentum anomalies increases with IVOL. • Irrespective of the level of IVOL, momentum should weaken or disappear after adjusting for information risk.
Agenda Ahead • Introduce ECIN • Construct ECINF from ECIN • Show that EICNF is a priced factor through two-stage-cross-sectional-regression (2SCSR) test • Test the information risk hypothesis of momentum • Extension
Motivating ECIN • Price discovery of trades: private information is revealed in the sequence of trade prices. • Glosten and Milgrom (1985) and Kyle (1985): private information revealed by the informed trading. • Informed traders prefer to trade in medium to large size • Easley and O'Hara (1987) ,Barclay and Warner (1993) ,Battalio and Mendenhall (2005) and Hvidkjaer (2006). • Stocks whose large trade have larger price discovery will have higher information risk. • ECIN is the price discovery of large trades.
Estimating ECIN • Large and small trade prices are co-integrated since they are the prices of the same stock. • Co-integrated series can then be represented by a specific form of the Vector Error Correction Model (VECM) • the price discovery of trades can then be estimated as the error-correction coefficient from this VECM Harris et al. (1995), Hasbrouck (1995) , Eun and Sabherwal (2003) , Werner and Kleidon (1996) use VECM to study the price discovery in different markets
PL and PS are non-stationary I(1) but they are cointegrated. is stationary process with zero mean. Estimating ECIN (II) PL PL αLZt-1 αLZt-1 Zt-1 Zt-1 αSZt-1 αSZt-1 PS PS t-1 t Zt-1>0 Zt=0 Large Price Discovery More Liquid t-1 t Zt-1>0 Zt=0 Small Price Discovery Less Liquid
Data Samples • NYSE and Amex • January 1983-December 2006 for ECIN estimation • January 1984-December 2007 for asset pricing test • VECM is estimated for each firm in each calendar year
Large and Small Trade Classification • (1) For each stock i and in each month m, we sort individual trades by their dollar-based trade size. We then identify the 60th percentile of the dollar-based trade size for stock i in month m. • (2) The average of the monthly 60th percentiles over every five-year period beginning in 1983 are used as the criterion for the dollar-based trade size for stock i for that particular period. Say it is $20,000. • (3) We divide the dollar-based criterion from Step (2) by the closing price in month t-1 of stock i to obtain the number of round lot shares. This number of round lot shares is the share-based trade size criterion for stock i in month t. For example, if the price of stock i in 198506 is $10, then the share-based trade size cutoff is 20,000/10=2000 shares in 198507.
Prices Series of Large and Small Trades • MINISPAN (Harris, McInish, Shoesmith, and Wood (1995)) • Retrieve the matched pairs at fixed intervals of 20 minutes. S L S L S S S L S S L • Open and close trades are deleted
Different from liquidity effect • The strong liquidity effect dominates the positive ECIN in January. • High ECIN firms are more liquid, so liquidity effect weaken the ECIN effect, but the positive ECIN effect dominates the negative liquidity effect. • The liquidity effect is strongest in January due to tax loss selling
Creating the ECINF Factor Market Cap
Creating the ECINF Factor Stock Returns ECIN portfolio=(HECIN-LECIN)-(SSIZE-LSIZE)
Momentum and Information Risk • H(1) Losers have lower information risk than Winners do. • H(2) After adjusting for the information risk differential between Winners and Losers, the momentum anomalies would weaken or disappear altogether.
Momentum and Idiosyncratic Volatility • H(3) The information risk decreases with idiosyncratic volatility for both good news and bad news firms. • H(4) The information risk differential between good news and bad news firms, and hence the momentum anomalies increase with idiosyncratic volatility.
Extension: Other Anomalies Explained by ECINF • SEO underperformance /Equity issuing puzzle /Asset growth anomaly • Lower information risk after issuing equity, SEO and asset growth • Idiosyncratic Volatility (IVOL) Puzzle/Dispersion Effect • High IVOL firms and firms with high disperison in A • Accrual Anomaly • High Accrual firms have lower information risk.
Conclusions • Construct an information risk factor ECINF and show it is is a priced factor. • ECINF fully explains both Price momentum and PEAD, because losers (bad news firms) have lower information risk. • ECINF also explains why momentum anomalies are stronger when information uncertainty (volatility) is higher. • ECINF can also explain a wide range of other anomalies