The Integrity of Financial Analysts: Evidence from Asymmetric Responses to Earnings Surprises

Source: the CFA Institute Financial Market Integrity Outlook Survey (http://​www.​cfainstitute.​org/​ethics/​topics/​Pages/​financial_​market_​integrity_​index.​aspx).

In contrast, buy-side analysts are employed by pension-fund or mutual-fund companies and manage money on behalf of their clients. These analysts research stocks and make recommendations to the funds’ financial managers. Conflicts of interest are generally of less concern among buy-side analysts.

Prior studies report that large earnings surprises, particularly large negative earnings surprises, are costly to firms (e.g., Mikhail et al. 2004; Doyle et al. 2006; Ng 2007). Managers are thus motivated to avoid large negative earnings surprises and report earnings that are consistent with market expectations (e.g., Kasznik and NcNichols 2002; Matsumoto 2002). Firms facing large negative earnings surprises are also more likely to make discretionary disclosures to warn investors about disappointing earnings (Kasznik and Lev 1995).

In comparison with earnings that meet or marginally exceed analysts’ expectations, which many researchers interpret as the outcome of earnings management (e.g., Burgstahler and Dichev 1997; Hayn 1995), large positive earnings surprises are less likely to be the result of managers’ earnings manipulation. Likewise, large negative earnings surprises are less likely to be the result of firms’ use of the “big bath” technique. Large negative earnings surprises may be interpreted as an indication of a firm’s financial distress because it is relatively difficult for management to boost earnings through earnings management to an extent that they can substantially meet analysts’ earnings expectations.

See the news release by the Securities and Exchange Commission (SEC) at http://​www.​sec.​gov/​news/​press/​2003-54.​htm.

The CFA Institute, a global association of professional financial analysts, recently published a Code of Ethics and Standards of Professional Conduct (effective from July 1, 2014), which defines principles that help analysts to manage their conflicts of interest.

We collected analysts’ recommendation ratings and earnings surprises from Yahoo! Finance (http://​finance.​yahoo.​com). The information on the firm’s securities issuance was drawn from its 2006 10Q form.

Responses may also be compromised merely by analysts’ ignorance or lack of research. We indirectly investigate these possibilities by adding analyst experience as a control variable in our regression analysis.

Previous researchers also scale the difference between I/B/E/S earnings per share and analysts’ consensus earnings forecasts by assets per share (Core et al. 2006), the standard deviation of earnings forecasts (Mendenhall 2003), and market price per share at the beginning (or end) of quarter q (Franzoni and Marin 2006). We use each of these measures of earnings surprises in our robustness checks and obtain similar results. Another measure of earnings surprises is standardized unexpected earnings. This measure is predicated on the assumption that earnings follow a seasonal random walk model with a drift and is commonly used in the literature on postearnings announcement drift. However, as the focus of the current study is analysts’ reactions to unexpected earnings, we measure earnings surprises relative to analysts’ forecasts rather than using a time-series model of firms’ prior earnings.

In simple terms, we focus on the 30 % of earnings surprises with the largest absolute values.

Conrad et al. (2006) assume that an investment banking relationship exists if any debt, IPO, SEO, or M&A transaction is conducted by the analysts’ firm at any time during the sample period. We repeat our tests using this definition of affiliated analysts and obtain results similar to those presented here.

As presented in Fig. 1, this recommendation may be made either after the firm’s quarter q fiscal period (illustrated by a solid line) or during the quarter q fiscal period (illustrated by a dotted line). We make sure that this recommendation is before the next quarterly EAD.

Unlike some prior studies, we select the earliest stock report following the earnings announcement for quarter q rather than the most recent forecast report for the next quarter, q + 1, for the following reasons. First, approximately 26.9 % of the recommendations in our sample are made within 7 trading days (i.e., the EAD plus the next 6 trading days) of the firms’ announcements of their quarterly earnings news. Another 37.2 % of the recommendations are made between the next 8th and 15th trading days. Collectively, more than 60 % of the recommendations are made within 15 trading days of the EAD. Therefore, the earliest report reflects analysts’ immediate response to the arrival of new information and is most relevant to our study. Second, forecast immediacy, or the speed with which analysts respond to a significant change in publicly available information, is positively related to forecast usefulness (Mozes 2003). Third, as the focus of this study is analysts’ responses to large earnings surprises reported in the previous quarter, we need to control for changes made to analysts’ recommendations in response to important firm information other than quarterly earnings announcements. Using the earliest forecast report minimizes the effects of other information on analysts’ recommendation changes in the time window between large earnings surprises and subsequent analysts’ recommendation changes.

In particular, as discussed by Kadan et al. (2009), Kolasinski and Kothari (2008), and Wu et al. (2015), not all brokerages use a 5-tier recommendation system. Prior to the Global Settlement, about 17 % of recommendations were issued using a 3-tier (buy/hold/sell) system, and this proportion rose to over 75 % following the Global Settlement. Although I/B/E/S has coded the recommendation levels at a 5-tier system, the calculated changes in recommendation levels across brokerages and analysts using different tier-systems are not comparable. We are grateful to one of the referees for sharing the insight regarding this issue.

A shortcoming of this measure of analyst experience is that it does not accommodate analysts’ research reports before October 1993, the first month for which I/B/E/S recommendation data are available. An alternative measure is to count analysts’ research reports only after a specific year (e.g., 1995). We use this measure as a robustness check and obtain similar results on this issue.

Our institutional-holdings data are drawn from Thomson Financial/Spectrum.

This list of prestigious investment banks is similar to Fang’s (2005) list, which comprises Goldman Sachs, Merrill Lynch, Morgan Stanley, Salomon Brothers, Credit Suisse, Lehman Brothers, JP Morgan, and DLJ. Note that in August 2000, Credit Suisse acquired DLJ. We add three banks (UBS, Barclay Capital, and Citi) to the list based on a recent ranking of investment banks by The Wall Street Journal (October 1, 2009). Our results do not change significantly if the top 8 or top 15 banks are selected as prestigious banks.

Jay Ritter’s ranking is primarily based on the CM system and can be accessed at http://​bear.​warrington.​ufl.​edu/​ritter/​ipodata.​htm.

We also exclude utilities (Standard Industrial Classification (SIC) codes between 4400 and 4499) and financial institutions (SIC codes between 6000 and 6999).

The industry-adjusted ROE is equal to net income before extraordinary items divided by book value of equity and adjusted by industry median ROE. The industry classification is based on Fama–French’s (1993) system.

As the distribution of analyst recommendations is non-normal and right-skewed, we report bootstrapped p-values rather than conventional t-statistics. Following Hertzel et al. (2002), we perform the bootstrapping procedure as follows. First, we calculate the average recommendation levels for affiliated analysts in the large positive earnings surprises sample before and after the quarterly EAD and obtain the difference between them (REC_after − REC_before). We then group the recommendation ratings and randomly select recommendation ratings with replacements to construct our first pseudosample. Next, we estimate the recommendation change for this pseudosample as the first mean-difference observation (recommendation change). We repeat this procedure 1000 times to obtain 1000 observations of pseudosample recommendation changes. This procedure yields empirical distributions of recommendation changes under the null hypothesis of no mean difference. Finally, the null hypothesis is rejected at the α % level if the recommendation change for our sample firms is less than the (1 − α) percentile recommendation changes in the empirical distribution of the pseudosamples. We apply the same procedure to the large negative and moderate earnings surprises samples.

The interpretation of interaction effects in nonlinear models (such as a logistic regression used here) is not quite as simple as in linear models. A significant coefficient for an interaction is not necessarily evidence of a significant difference in probabilities across groups. Therefore, following a comment from one of our anonymous referees, we report the difference in marginal effects across groups.

Regarding the probability of no change of recommendation (UPGRADE = 0) in response to large negative earnings surprises, the marginal effect of analyst affiliation status (AFFIL) for earnings surprises is also significantly more positive for affiliated analysts than for unaffiliated analysts (difference in marginal effects = 0.036, p < 0.001).

Under the Global Settlement, 10 of the largest Wall Street banks paid $1.4 billion to federal regulators to settle the charge made by the government that the banks had issued optimistic stock reports to win investment banking clients. Jack Grubman, once a top analyst at Salomon Smith Barney, paid millions in fines and was banned from the investment industry for life. The involvement of highly regarded analysts and banks in the scandal appears to support our findings.

Due to space limitations, this table is not presented here. It is available from the authors upon request.

We thank one of our referees for these suggestions. The results are not tabulated here to save space; they are available from the authors upon request.

We include stock returns 1 day before the issuance of the analyst’s recommendation report to incorporate possible information leakage.

The mean-adjusted abnormal volume is much greater in reaction to unaffiliated upgrade announcements than to affiliated upgrade announcements.