Analyst recommendations and the implied cost of equity

Sell-side financial analysts make public recommendations about the prospects of publicly traded companies. In particular, they make recommendations to buy, hold, or sell the equity stock of specific companies, and they change their recommendations from time to time. Anecdotally, the market price effect of recommendations could largely depend on the number of investors adopting such recommendations and the quality and completeness of information contained in the recommendations.

Consistent with this idea, Zhou (2013) finds a positive association between recommendations and target prices levels.

Kecskes et al. (2017) document that 46.5% of their sample of recommendation changes is associated with concurrent earnings forecast changes announced sometime during the same week.

Asquith et al. (2005) report that only 12.8% of analyst reports include some variant of discounted cash flow used for firm valuation, which would necessitate the use a discount rate.

Price targets are included in 72.6% of analyst reports according to Asquith et al. (2005).

For example, Aggarwal et al. (1995) show that analysts do not use all available information to improve their forecasts when forecasting announced values of macroeconomic variables. Elgers et al. (2016) conclude that predictive ability of analyst forecasts is time specific, such that forecasts are less accurate after Regulation DF.

We exclude years during and following the subprime mortgage crisis to avoid biasing our results. Research on analyst recommendations suggests that recommendations are less informative during periods of financial turmoil (Barber et al. 2003; Woolridge 2004).

For robustness, we also repeat our analysis using ICOE estimates of the individual models. The results are qualitatively consistent with what we report based on the average, with some variation in the strength of this association across models.

The possibility of measurement error in ICOE could lead to spurious results relating ICOE and analyst recommendations. By using the average of the four models, we reduce errors that may occur with any one of the models. We also repeat our analysis using each individual model and the results are consistent. Apart from this, we use the changes in the cost of equity, which helps mitigate measurement errors in ICOE that persist through time.

These models and their description, to a large extent, resemble those in other papers such as Attig et al. (2008) and Dhaliwal et al. (2006).

We also use K to represent the cost of equity risk premium for the final variable construction by subtracting the 10-year Treasury bond yield. For convenience, we continue to call this “cost of equity.”

Dividend payout used with the Claus and Thomas (2001) model is fixed at 50% for all firms, consistent with Claus and Thomas (2001).

For robustness, we repeat the analysis using changes in mean recommendations. The results are qualitatively unchanged.

If recommendations were informative, then we would expect a negative relation between past recommendation changes and the current cost of equity changes (ICOE should decrease following an upgrade and increase following a downgrade), or no relation if the market had sufficient time to incorporate the information before the current period started. On the other hand, an appropriate analyst response predicts a positive relation between past cost of equity changes and current recommendations (analysts should upgrade following an increase in ICOE and downgrade following a decrease), or no relation if the analysts had sufficient time to respond before the current period started.

Brown et al. (2014) find that abnormal returns have a strong reversal especially when mutual fund managers herd following analyst changes. They interpret this as an overreaction to analyst recommendation revisions and subsequent reversion. However, much of the abnormal return occurs several quarters prior to the recommendation revision event, so their results are also consistent with analysts belatedly changing their recommendations after the fact. We investigate this issue in Models 3, 5, and 7. Cremers et al. (2017) show that positive analyst recommendations correspond to short-term market overreactions and subsequent reversals for stocks frequently traded by short-term institutional investors that are difficult to arbitrage.

For most of the firms (over 88% of sample firms) mean or median target price estimate is higher than the market price on the statistics release date for our cost of equity estimation month. On average analyst recommended target prices are about 25% higher than the actual closing market price on that day. This finding is not new, as the literature overwhelmingly shows that analysts are overly optimistic (Asquith et al. 2005; Brav and Lehavy 2003).

Jegadeesh et al. (2004) primarily investigate how various factors and firm characteristics (called Quantitative Score factors) affect how recommendations and recommendation changes predict future returns. They also look into how Q-Score factors affect recommendations and recommendation changes. While we also investigate how various factors affect recommendation changes, our primary concern is how ICOE affects recommendation changes, which is not considered by Jegadeesh et al., while controlling for other factors that affect recommendation changes.

Brav and Lehavy (2003) also find that the ICOE provides significant information beyond that provided by recommendation changes: They find a positive association with the abnormal return around the revision event.

Ho (2000) finds that downgrades are more informative than upgrades.

Our proxy for earnings quality is estimated following the method applied in Francis et al. (2005) using the model of Dechow and Dichev (2002). The regression model that explains the total current accruals for firm j in year t (TCA _j,t ) is given by

where TCA _j,t  = ΔCA _j,t  − ΔCL _j,t  − ΔCASH _j,t  + ΔSTDEBT _j,t , CFO _j,t  = NIBEj,t − TA _j,t is firm j’s cash flow from operations in year t, NIBE _j,t is firm j’s net income before extraordinary items in year t (Compustat Data #18), TA _j,t  = ΔCA _j,t  − ΔCL _j,t  − ΔCASH _j,t  + ΔSTDEBT _j,t  − DEPN _j,t is firm j’s total accruals in year t, ΔCA _j,t is firm j’s change in current assets (Compustat Data #4) between year t − 1 and year t, ΔCL _j,t is firm j’s change in current liabilities (Compustat Data #5) between year t − 1 and year t, ΔCASH _j,t is firm j’s change in cash (Compustat Data#1) between year t − 1 and year t, ΔSTDEBT _j,t is firm j’s change in short-term debt (Compustat Data #34) between year t − 1 and year t, DEPN _j,t is firm j’s depreciation and amortization expense (Compustat Data #14) in year t, ΔREV _j,t is firm j’s change in revenue (Compustat Data #12) between year t − 1 and year t, PPE _j,t is firm j’s gross value of property, plant and equipment (Compustat Data #7) in year t. We estimate the above equation for the firms in each Fama and French (1997) 48 industry group that has a minimum of 20 firms in the sample. (We also winsorize the data at the 1 and 99 percentiles.) For each firm, the earnings quality metric is the standard deviation of the residuals for 5 years before and including the estimation year.