The dispersion anomaly and analyst recommendations

Theoretically, there should be a positive relationship between analyst forecast dispersion and expected stock returns. Cragg and Malkiel (1982) and Givoly and Lakonishok (1984) argue that the forecast dispersion should be priced as a risk since it captures information asymmetries among the investors. Differently, Varian (1985) argues that outperformance of high-dispersion stocks is a result of lower current prices for stocks with high dispersion due to higher sales made by pessimistic investors. However, empirical support of the priced risk story is very limited (Qu et al. 2003; Barron et al. 2009).

Cowles (1933), Bidwell (1977) and Diefenbach (1972) find that on average analysts’ recommendations do not produce abnormal returns, a result consistent with efficient market hypothesis. On the contrary, Davies and Canes (1978), Liu et al. (1990), Beneish (1991), Barber and Loeffler (1993) and Liang (1999) find that the “buy”/”sell” recommendations published in Wall Street Journal’s “Heard on the Street” and “Dartboard” columns result in abnormal returns and prices. Womack (1996) documents significant positive (negative) persistent abnormal returns associated with “buy” (“sell”) recommendations. Barber et al. (2001) find that portfolios formed using the most favorable consensus analyst recommendations generate on average annual abnormal gross return of 4.13% after controlling for other risk factors, which dissipates once they account for transaction costs. More recently, Palmon et al (2009) document positive abnormal returns surrounding columnists’ stock recommendations, especially if these recommendations refer to managements officials or contain merger and acquisition news.

For example, Francis and Philbrick (1993), find that analysts’ earnings forecasts become more optimistic as the stock recommendations become less favorable, possibly due to analysts wanting to curry favor with management in order to gain greater private information. On the contrary, Eames et al. (2002), find that analysts issue more optimistic (pessimistic) forecasts for “buy” (“sell”) than for “hold” stocks in order to stimulate stock trading. Bradshaw (2004) finds that analysts’ projections of long-term earnings growth have the greatest explanatory power for stock recommendations, but investment strategies based on these projections have the least association with future excess returns, suggesting that analysts’ recommendations are based on ad hoc heuristics.

I/B/E/S reports analysts EPS forecasts starting from 1976, but with an increased number of forecasts starting from February 1982 and analyst recommendations starting from November 1993.

Market value of equity is computed as the product of the price per share (PRCC_F) at the end of the fiscal year times the shares outstanding (CSHO). Book value of equity is total assets (AT) minus total liabilities (LT) plus deferred taxes and investment tax credit (TXDITC) minus book value of preferred stock (in the following order: redemption value (PSTKRV) or liquidating value (PSTKL) or carrying value (PRTK). Book-to-Market ratio is the ratio between the book value of equity and the market value of equity as define previously. The book-to-market ratio for returns from July of year t to June of year t + 1 is constructed using accounting data at the end of year t − 1.

http://​mba.​tuck.​dartmouth.​edu/​pages/​faculty/​ken.​french/​data_​library.​html.

The Unadjusted Detail History file contains individual analysts’ forecasts organized by the date on which the forecast was issued. Each record also contains the revision date, the date on which the forecast was last confirmed as accurate. Similar to Diether et al. (2002), I extend each forecast until the revision date. Say, if the forecast was made in May and was last confirmed as accurate in July, it is used to compute mean and standard deviation forecast as well as analyst coverage for May, June and July.

The use of this measure is limited due to the data available, and caution should be exerted when interpreting the empirical results.

For example, Miller and Scholes (1982), argue that the cross-sectional relation between the bid-ask spread and stock returns could be spurious and driven by the stock price.

See: Amihud and Mendelson (1986), Amihud (2002), Spiegel and Wang (2005) for such evidence.

They develop a bid-ask spread estimator from daily high (ask) and low (bid) prices, given the assumption that daily high (low) prices represent buy (sell) orders. Apart from being easy to construct, the high–low spread estimator proposed by Corwin and Schultz (2011) is highly correlated to TAQ effective spreads (0.87 for monthly spreads and 0.75 for weekly spreads).

See: Merton (1987), Malkiel and Xu (2002), Barberis and Huang (2001) and Ewens et al. (2013) for such arguments.

See: Spiegel and Wang (2005), Eiling (2006), Brockman et al. (2007), Fu (2009) and Vidal-Garcia et al. (2016). However, studies such as Fink et al. (2012) and Guo et al. (2014) argue that EGARCH methods similar to Fu (2009) incorporate contemporaneous information in the estimated conditional variance, which contributes to the positive relationship between idiosyncratic volatility and expected returns. The relationship is not robust, once they control for this information.

Statistics for the rest of the portfolios are not reported. Results are available upon request.

Several portfolio analyses are performed on different samples including the whole sample 1983:02–2012:12, Diether’s et al (2002) sub-sample 1983:02–2000:12, sub-sample 2001:01–2012:12. The results are robust, documenting a strong negative relationship between the next-period stock returns and forecast dispersion, which is stronger for the whole sample. Moreover, similarly to Diether et al. (2002), I document a weaker dispersion-return relationship over the later period, that could be due to lower transactions costs and more accessible information in recent decades. These results are available upon request.

Results from the dependent double sort are similar to those of the independent double sort, and are available upon request.

Early studies on short interests by Figlewski (1981), Brent et al.(1990), Figlewski and Webb (1993), Woolridge and Dickinson (1994) do not find a strong relationship between short interest and excess returns, mainly due to sample selection based on the magnitude of the short interests. Asquith and Meulbroek (1995) suggest that firms that have very small short positions (less than 0.5%) are likely to represent hedge positions, rather than a systematic attempt to exploit the perceived overpricing.

Dechow et al. (2001) suggest that investors may use short-selling in a merger situation, “pairs trading”, or to arbitrage a price differential between the stock and debt convertible to stock. In these situations the short selling of a stock is not motivated by the expected decline in the future price. However, Lee (2016) finds evidence that short sellers try to exploit temporary mispricing in stocks that is not related to illiquidity or short sale constraints.

Compustat Monthly Updates—Supplemental Short Interest File reports short positions since 1973. Short Interest Ratio is constructed using Boehme et al. (2006) approach. In similar fashion, firms for which short positions are not reported in the file are not included in the analysis, because there is not enough information on whether these firms face short sale constraints or simply do not report the short positions. The restricted dataset has a total of 96,016 firm-month observations.

In all tests, I employ double sorting for unadjusted returns and risk-adjusted returns using CAPM, FF3 and FF4. The tables are not reported in order not to overcrowd the paper. Results are available upon request.

I also employ triple sorting for unadjusted returns and risk-adjusted returns using CAPM, FF3 and FF4. The tables are not reported in order not to overcrowd the paper. Stronger results are reported for risk-adjusted returns using CAPM and FF3 models. For risk-adjusted returns using FF4 and FF5 results are similar. Results are available upon request.

Separate robustness regressions are conducted for subsamples of stocks with low and high short interest ratio. Stronger results are achieved for the group of stocks with high short sale constraints. Tables are not included in order not to overcrowd the paper. Results are available upon request.