Chapter 2. Literature 7'eview
2.4. Analysts' Earnings Forecasts
2.4.1. Usefulness o f analysts' earnings forecasts in accounting research
Since the underlying principle that the share price of a firm is the embodiment of the market's expectations about its future prospects seems to be true, knowing and quantifying market expectations is one of the most important factors in the investment process (Rosen, 2000). In this context, the quantification of market's expectations has long been a main concern of both practitioners and capital market researchers. Following the development of databases that collect and process brokerage earnings estimates (i.e., I/B/E/S, First Call, Value Line, Zacks), it has been getting much easier
16 Value Line database is preferred to other sources because it provides a more comprehensive set o f forecasted attributes over longer horizons (Francis et al., 2000, p. 51).
for both practitioners and researchers to access analysts' earnings forecasts for investment decisions and research. The usefulness and the attributes of analysts' earnings forecasts have been widely studied for the last three decades,17 and this section reviews some of these studies in the context of equity valuation.
As mentioned in the above section, even for the LID-based equity valuation research, analysts' earnings forecasts can play an important role as an input when defining the 'other information' variable in the linear information dynamics. Recall that analysts' earnings forecasts can be used directly in the EBO-type models. Thus, analysts' earnings forecasts appear as one of crucial components in many valuation models. In addition with U.S. evidence, some European studies have found that forecasts of future earnings are an important factor in equity valuation (Capstaff et al., 2001).
Earlier studies about analyst-based earnings forecasts examine whether analysts' earnings forecasts are superior to time-series model-based earnings forecasts that rely solely on past information. Many studies show evidence that analysts' earnings forecasts are more accurate than time-series forecasts (Brown and Rozeff, 1978; Fried and Givoly, 1982; Brown et al., 1987b; Brown, 1993). Capstaff et al. (1995) also provides similar results for the U.K. Brown et al. (1987b) show that the superior accuracy of analysts' forecasts over time-series forecasts is not an artifact of i) chronological subperiods, ii) forecast horizon, iii) forecast error definition or treatment of outliers, iv)
171/B/E/S Research Bibliography: Sixth Edition, edited by Lawrence Brown in 2000, consists o f 579 studies related to analyst expectations.
Chapter 2. Literature review
conditioning quarter, or v) the statistical test statistic on which inferences are drawn.18 Instead, Brown et al. (1987b) contend that the superior accuracy of analysts' forecasts is attributable to i) better utilization of information existing at the forecast initiation date for the time-series models (termed as a contemporaneous advantage), and ii) use of information acquired after the time-series model's forecast initiation date (termed as a timing advantage). In other words, the higher accuracy of analysts' forecasts over time- series forecasts generally stems from analysts' broader information set.
In terms of explainability of stock returns, Brown et al. (1987a) indicate that analyst- based earnings forecasts are generally more highly associated with abnormal stock returns than various time-series model-based earnings forecasts. However, O'Brien (1988) gives contradictory findings. Although her results are consistent with the higher accuracy of analysts' forecasts over time-series forecasts, she finds that autoregressive model forecasts explain abnormal stock returns better than analysts' forecasts. Despite some conflicting evidence on the accuracy and the explainability of analysts' earnings forecasts, it is common practice to implicitly assume that analyst-based earnings forecasts are a better surrogate for market's expectations than time-series model-based earnings forecasts (Kothari, 2001).
18 There is some extant literature that show findings against the superiority o f analysts' forecasts over time-series forecasts. These conflicting findings have led some researchers to attribute the superiority o f analysts' forecasts to an artifact o f certain experimental design issues (Brown et al., 1987b).
2.4.2. Attributes o f analysts' earnings forecasts
Evidence o f optimism
Besides studies that demonstrate the higher accuracy and explainability of analysts' forecasts over time-series forecasts, many researchers have also studied the attributes of analysts' earnings forecasts. An important question related to the properties o f analysts' earnings forecasts is whether analysts overestimate or underestimate earnings in a systematic way. This is the question about the bias of analysts' earnings estimates. If there is a systematic positive (negative) difference between forecasts and actuals, we call it as optimism (pessimism).
Notwithstanding the research design differences,19 numerous past academic studies provide evidence that analysts tend to be optimistic and the optimistic bias is evident for most years and forecast horizons (O'Brien, 1988; De Bondt and Thaler, 1990; Brown, 1997; Brown, 1998; Richardson et al., 1999; Easterwood and Nutt, 1999). Some studies even argue that analysts seem to be too optimistic for investors to rely on their forecasts (Dreman and Berry, 1995; Chopra, 1998). Chopra (1998) finds that the average earnings growth forecast is more than twice the actual growth rate. However, the median bias seems to be quite small or unbiased, indicating that the extreme outliers hugely influence on measures of optimism (O'Brien, 1988; Abarbanell and Lehavy, 2000a).
19 Research design across studies can be different in terms o f the use o f earnings forecasts (source o f forecasts (e.g. I/B/E/S or Value Line), median or mean, consensus or individual, quarterly or yearly, first or last), the use o f actual earnings (same or different source with forecasts), and/or the treatment o f outliers (trimming or winsorising, scaled or non-scaled forecast errors).
Chapter 2. Literature review
Determinants o f optimism
There are several possible explanations for the presence of an optimistic bias in analysts' forecasts of earnings per share. These explanations generally fall in two categories. One is economic incentives-based explanations, and the other is behavioral cognitive-bias explanations. Among incentives-based explanations, some possible reasons why analysts tend to bias their true predictions toward a more optimistic view could be because i) analysts fear jeopardizing potential investment banking business, ii) analysts fear losing access to management as a source of information, and/or iii) analysts seek to generate trading commissions (McNichols and O'Brien, 1997). Consistent with these explanations, affiliated analysts seem to forecast earnings more optimistically than unaffiliated analysts, and managers tend to penalize analysts who produce negative reports about their firms by limiting or cutting off analysts' future contact with them (Das et al., 1998).
On the other hand, De Bondt and Thaler (1990) and Capstaff et al. (2001) propose a cognitive-bias explanation for analysts' forecast optimism. They argue that analysts systematically over-react to new earnings information and the overreaction results in the optimistic forecasts. On the contrary, Abarbanell and Bernard (1992) find evidence that analysts under-react to earnings information, which is consistent with the post-eamings- announcement drift phenomenon. Recent research by Easterwood and Nutt (1999) shows that analysts underreact to negative information and overreact to positive information. They argue that these findings are consistent with analysts' optimism and bring together the apparently disparate conclusions of De Bondt and Thaler (1990) and Abarbanell and Bernard (1992).
McNichols and O'Brien (1997) provide another explanation for analysts' optimism. They argue that analysts choose not to publish unfavourable forecasts. That is, some portion of the pervasive analysts' optimism is due to analysts' self-selection of stocks for the coverage. Herding behavior among analysts could be one of other explanations for analysts' optimism as well (Brown, 1998).
Revisions
A deterrent to analysts from issuing optimistic forecasts could be the trade-off with their reputation. Optimistic analysts could be compensated by their employers and/or their targeting firms, but they may face difficulty of losing their reputation and job if they keep issuing incorrect forecasts to investors. Mikhail et al. (1999) provides evidence that there is a significantly negative relationship between analyst turnover and relative (not absolute) forecast accuracy. This issue is related to forecast revisions.
Bartov et al. (2000), Kasznik and McNichols (1999) and Lopez and Rees (2001) address that the meeting or beating earnings expectations (MBE) phenomenon is partly due to firms' earnings and expectation management, and the rewards to MBE are significant. Thus, even though firms' general management of analysts' expectations leads analysts to issue favorable (i.e., optimistic) forecasts, the expectation management as the fiscal year end approaches seems to encourage analysts to closely meet the following actual earnings because of MBE phenomenon. From this point of view, analysts are likely to revise their forecast downward in order to get credibility from the market. Chopra (1998) and Richardson et al. (1999) also report that analysts' forecasts
Chapter 2. Literature review
are revised downward continuously in the course of the year.
Decline in optimism
Analysts' optimism appears to be waning in recent years. Brown (1997) shows that analysts' optimistic bias has decreased over time and was absent for S&P 500 firms from 1993 to 1996. Brown (1998) and Richardson et al. (1999) also provide evidence that the bias has turned from optimism to pessimism in recent years.
Kothari (2001) summarises three hypotheses that are possibly consistent with the decline in analysts' optimism. First, as Clement (1999), Jacob et al. (1999) and Mikhail et al. (1999) documented in their studies, analysts' experience seems to be positively
associated with forecast accuracy. That is, by learning from past biases, analysts could reduce their optimistic bias. Second, as shown above, analysts' incentives may have changed. Because MBE is likely to be strongly associated with stock price, analysts have incentives to alter their initial optimistic forecasts to the most plausible figures as the earnings announcement date approaches. Third, the quality of data used in the research examining analysts' forecast properties has improved. Abarbanell and Lehavy (2000b) argue that the development of earnings definition that forecast data providers require analysts to forecast is a main factor in explaining the recent declines in analysts' optimism.