SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. 1 Introduction
  4. 2 Background and research questions
  5. 3 Data and sample
  6. 4 Research design
  7. 5 Test results and discussion
  8. 6 Concluding remarks
  9. Appendix
  10. References

Using an extensive database of 356,463 sell-side equity analysts' reports from 2002 to 2009, this study is one of the first to analyze the readability of analysts' reports. We first examine the determinants of variations in analyst report readability. Using several proxies for ability, we show that reports are more readable when issued by analysts with higher ability. Second, we test the relation between analysts' report readability and stock trading volume reactions. We find that trading volume reactions increase with the readability of analysts' text, consistent with theoretical models that predict that more precise information (and hence more informative signals) results in investors' initiating trades. These results support the view that the readability of analysts' reports is important to analysts and capital market participants.


1 Introduction

  1. Top of page
  2. Abstract
  3. 1 Introduction
  4. 2 Background and research questions
  5. 3 Data and sample
  6. 4 Research design
  7. 5 Test results and discussion
  8. 6 Concluding remarks
  9. Appendix
  10. References

We investigate the importance of analyst report readability. Analysts' reports provide investors with relevant firm and industry information.1 Higher readability of these reports can lower investors' cost of acquiring information by decreasing the time spent to understand the reports. Apart from the incentive to attract investors, the readability of these reports is also influenced by equity analysts' incentives to generate trading volume. These incentives include trading commissions, which provide an important source of revenues for the brokerage firms that employ the analysts. Consistent with economic theory and conditioning on the information content of the report, we expect analysts to generate greater trading volume by providing more precise written information (i.e., more readable text) in their reports to investors; as the more precise information is expected to result in investors initiating trades (e.g., Karpoff 1987; Kim and Verrecchia 1991). Given analysts' incentives to serve their clients and to generate trading commissions, the importance of producing readable reports becomes an empirical question.

We first examine the determinants of analyst report readability. Specifically, we investigate whether readability is associated with “high-ability” analysts, broadly defined to include those analysts who have more experience, issue timelier forecasts, forecast more frequently, are ranked by Institutional Investor magazine (hereafter, II), and issue more consistent earnings forecasts and stock recommendations. Second, we investigate whether more readable analysts' reports lead to greater trading volume.

Our tests use an extensive database of 356,463 analysts' reports downloaded from Investext for U.S. public companies from 2002 to 2009. We create and study measures for two aspects of readability: straightforward language and concise reports. To measure the former, we aggregate three commonly used measures of readability: the Fog, Flesch, and Flesch-Kincaid indices. To measure the latter, we follow Li (2008) and use the number of words and the number of characters in the report and aggregate these two.

We find that more readable reports are issued by analysts who have more experience, issue more timely forecasts, revise forecasts more frequently, and are more likely to issue consistent forecast and recommendation revisions. These results support the notion that high-ability analysts provide more readable reports. Our tests include variables that measure task complexity both at the analyst level (number of industries followed) and at the firm level (stock return volatility, the inverse of firm age, and the number of business segments). As expected, we find that complexity is negatively related to readability. We also find that the number of companies an analyst covers is positively related to readability, which suggests a greater understanding of the industry followed by the analyst.

We next find that, after controlling for the determinants of analyst report readability (using two-stage least squares regressions), trading volume reactions are positively associated with the readability of analysts' text. This short-window event study effect is statistically significant, and the finding is consistent with theoretical models that predict that more precise information (and hence more informative signals) causes investors to initiate trades. The analyses control for the tone of the report. Our tests additionally control for firms' earnings and management forecast news, as well as other aspects of the reports such as revisions in the analyst's recommendations, earnings forecasts, and target prices. Finally, all empirical tests include controls for firm characteristics as well as broker, industry, and year fixed effects.2

We contribute to the understanding of analysts' communications by focusing on the readability of their text, which is not examined in the extensive literature on analysts' outputs (such as recommendations and forecasts). Second, we add to the burgeoning research that uses the tools of computational linguistics, which for analysts' reports is limited to Kothari, Li, and Short (2009) and Huang, Zang, and Zheng (2012) for equity analysts' reports and to De Franco, Vasvari, Vyas, and Wittenberg-Moerman (2013) for debt analysts' reports. Third, we respond to Bloomfield's (2008) call for business linguistics research that examines communications that are more spontaneous and timely than annual reports. Analyst reports are less likely to be “rehearsed” and edited than annual reports as they are typically issued soon after corporate events (such as earnings releases).

2 Background and research questions

  1. Top of page
  2. Abstract
  3. 1 Introduction
  4. 2 Background and research questions
  5. 3 Data and sample
  6. 4 Research design
  7. 5 Test results and discussion
  8. 6 Concluding remarks
  9. Appendix
  10. References

Prior research

A large body of research examines and relies on the outputs of analysts, particularly summary measures such as earnings forecasts and stock recommendations. However, compared with the huge literature on earnings forecasts and stock recommendations, research on the content of analysts' reports and notes is limited.

A few studies examine whether analysts' written details are informative. Asquith, Mikhail, and Au (2005) create a variable that measures analysts' arguments as stated in the text of reports. They find this variable to be incrementally informative to forecast, recommendation, and target price revisions for a sample of 636 reports. De Franco and Hope (2011) test whether analysts' notes provide new information for investors using a large sample of over 380,000 notes. Interestingly, when directly comparing notes and earnings forecasts, they find that notes and forecasts have the same estimated coefficients when using the absolute value of abnormal returns as the test metric and that notes have a larger estimated coefficient than forecasts when using trading volume as the metric of investor reaction. These findings indicate that analysts' written disclosures are economically important and provide new information to investors.

To quantify the quality of analysts' written disclosures, we follow a burgeoning literature that uses computational linguistics methods to analyze the text of various business documents in order to measure different constructs such as tone, ambiguity, or uncertainty.3 One such paper is Li (2008), which examines disclosure transparency of annual reports by examining the readability, or “Fog index,” of disclosures. His results indicate that firms' annual report readability captures management's obfuscation behavior.

Specific to analysts' disclosures, a new literature is emerging on the textual analysis of analysts' written reports. Studies such as Asquith et al. (2005), Previts, Bricker, Robinson, and Young (1994), and Rogers and Grant (1997) use relatively small samples and manually code the written content of analyst reports. More recently, Kothari et al. (2009) analyze the content of more than 100,000 disclosure reports by management, equity analysts, and the business press. They find that when content analysis of management and the business press indicates favorable (unfavorable) disclosures, the firm's risk declines (increases) significantly.4 Another aspect of analysts' written disclosure is studied in De Franco et al. (2013). They code the tone of debt analysts' discussions about events that generate debt–equity conflicts of interest and show that the tone explains the difference between debt and equity analysts' recommendations. Further, De Franco et al. show that the textual content of analysts' reports impact the market, and they demonstrate that bond trading volume and credit default swap spreads are increasing in debt analysts' negative discussions of conflict events.

The impact of qualitative or soft information on capital markets is also considered by other studies (e.g., Brochet, Narajanjo, and Yu 2012; Dambra, Wasley, and Wu 2013). For example, Huang et al. (2012) use a Naïve Bayesian machine learning approach to extract opinions from the text in analyst reports, and they find that the stock prices react significantly to textual opinions.

Analyst ability and report readability

We start by examining the determinants of variations in analyst report readability. Our analyst report setting seems more aligned with theories underlying lexical analysis that implicitly assume that communications are spontaneous and represent unconscious behaviors and thought processes (Bloomfield 2008). In his study of annual report readability, Li (2008) derives his hypotheses from Bloomfield's (2002) incomplete revelation hypothesis. But annual reports tend to be heavily rehearsed in contrast to analyst reports that are typically issued relatively quickly following corporate events. Accordingly, while we use Li's “obfuscation” explanation when we investigate analysts' investment banking ties, we believe that the readability of analysts' reports is less likely to be driven by analysts' incentives to obfuscate bad news released by the firm; rather it is more likely to be a function of analyst characteristics such as ability.

We begin with the prediction that, ceteris paribus, readability is positively related to the ability of the writer (in our case, the analyst). This hypothesis relies on the idea that innate ability (ability acquired through education or experience) and the specific ability to write more readable reports are likely to be jointly determined. Prior research has documented some of these positive associations between ability and performance—see for example, De Franco and Zhou (2009). Motivated by prior studies including Clement (1999), Clement and Tse (2003), and Ramnath, Rock, and Shane (2008), we examine whether “high-ability” analysts issue more readable reports.

Trading volume reaction to analyst report readability

We predict that greater report readability leads to increased trading volume. Analysts' written reports consist of objective or hard information, as well as more qualitative or soft information that describes the firms' business and economic environment and provides nonquantitative forecasts and guidance. Investors value analysts' explanations regarding the economic and business environment in which the firms covered by them operate (Bradshaw 2011). Such explanations are likely to be conveyed, not through summary forecasts and recommendations, but through discussions in written reports. As Brochet et al. (2012) point out, understanding these discussions requires the readers to process implicit communication signals. In such settings, the clarity of communication is essential. In the context of analyst reports, this line of reasoning leads us to argue that the readability of analyst reports influences investor judgment about the analyst's report and, accordingly, influences their trading behavior.

Our hypothesis has support from extant theoretical work that more precise information leads to greater trading volume (e.g., Karpoff 1987; Kim and Verrecchia 1991). The intuition behind these models is that more informative signals cause investors to initiate trades.5 Barron and Karpoff (2004) argue that in a majority of theoretical trading volume models, more precise information signals increase investors' confidence about their private valuations, making them more willing to take speculative positions, thereby increasing trading volume. Based on this reasoning, we argue that greater readability implies a more precise signal, which is likely to be more informative for investors, and thus more readable reports are likely to increase trading volume.6

With regards to readability as measured by the conciseness of reports, Caskey (2009) shows that ambiguity-averse investors may prefer aggregate information even at the cost of a loss in information. Therefore, we expect that concise (and hence, more readable) reports have greater trading volume reactions.

Although we provide arguments supported by extant published research to support a directional hypothesis, the relation between readability and trading volume reactions is not tautological. Specifically, an alternative stream of theoretical research supports the null that less readable analyst reports would entail greater trading volume. The idea here is that greater readability decreases divergence of opinion among investors, which in turn reduces the incentives to trade (e.g., Harris and Raviv 1993; Kim and Verrecchia 1994; Kandel and Pearson 1995).

Finally, a third alternative is that readability has no effect on trading volume. Summary measures (i.e., forecasts, recommendations, and target prices), should capture and reveal a large part of the analyst's new information (De Franco and Hope 2011). We control for earnings forecast revisions, stock recommendation revisions, and target price revisions in our multivariate tests. In addition, summary measures, in contrast to the text, represent “hard” information that is measurable and ex post verifiable (see also Brochet et al. 2012). As such, summary measures could provide more credible signals, and investors could put more weight on the summary measures than on the readability of the text.7 Further, we control for the tone of the report by including a measure of whether the tone of the report has become more or less positive.8

Based on the above discussion, it is an empirical question which of these forces will dominate, and we consequently test (without an ex ante directional prediction) whether the readability of analysts' reports affects trading volume, and if so whether this effect is incremental to numerous other factors that are likely to drive volume. However, a careful examination of the ambiguity concept used in the hypotheses above suggests that the hypothesis that is likely to dominate is the one that predicts a positive relation between readability and trading volume. The notion of ambiguity discussed in the Karpoff (1987) and Kim and Verrecchia (1991) models, which pertains to the incremental informativeness of the analysts' output signals and leads to an increase in trading volume for more readable reports, seems more directly related to the readability of analyst reports.9 In contrast, the ambiguity concept discussed in the Harris and Raviv (1993) and Kim and Verrecchia (1994) papers, which leads to a decrease in trading volume for more readable reports, is more likely to be manifested in the overall information set possessed by investors. Therefore, we are comfortable in contending that more readable reports are, ceteris paribus, more informative.

3 Data and sample

  1. Top of page
  2. Abstract
  3. 1 Introduction
  4. 2 Background and research questions
  5. 3 Data and sample
  6. 4 Research design
  7. 5 Test results and discussion
  8. 6 Concluding remarks
  9. Appendix
  10. References

We obtain our sample of equity analysts' reports from Investext for the period from 2002 to 2009. Before downloading any reports, we take the intersection of U.S. companies with sales and total asset data available on COMPUSTAT and monthly equity return data available on the Center for Research in Security Prices (CRSP) to create a potential sample of 8,086 unique firms. Firms must also have their equity securities traded on the NYSE, AMEX, or NASDAQ markets. We then search Investext for these 8,086 firms over this time period.

We obtain two types of data from Investext. We download analyst reports in a PDF format. We also extract the header information for each report. This includes the name of the analyst writing the report, the name of the broker employing the analyst who issues the report, the title of the report, the name of the firm analyzed in the report, the length in pages of the report, and the unique number assigned by Investext to that particular report.

We use the Investext header information to match Investext analyst names with I/B/E/S analyst names and verify these matched analyst-name pairs using broker names. This process produces 4,854 matched analyst-name pairs and 639,829 reports. We link companies being covered by these matched analysts-name pairs in both Investext and I/B/E/S using the Tickers, Committee on Uniform Security Identification Procedures (CUSIPs), and Company Names. Next, since we code these analyst-report files using computational linguistic programs, we convert the PDF files into text files using a Python program (in particular, the PyPDF library). These text files are then merged with the header information using the unique Investext report number. Last, we use a Python program to compute the computational linguistics-based measures of readability described below.10

The likelihood of successfully converting PDF files increases for more recent time periods, as many reports are scanned image files. This motivates us to include time fixed effects in our tests. Also, the ability to convert PDF files is partially related to which broker issues the report. Brokers' source documents are initially converted into PDF files for distribution to clients in different ways and using different word processing programs across brokers. This provides additional motivation to control for broker fixed effects in our tests (beyond the possibility that readability may be affected by broker-specific boilerplate language). Other than over time and across brokers, we have no reason to believe this sample reduction is correlated with any of our main test variables.

Last, we require all variables in our empirical models to be nonmissing. Our final sample comprises 356,463 Investext reports matched to I/B/E/S, and we use this sample across our tests (except when testing for Fcst-Rec Consistency and II-Rank). There are 4,014 unique firms and 2,334 distinct analysts in the final sample.

4 Research design

  1. Top of page
  2. Abstract
  3. 1 Introduction
  4. 2 Background and research questions
  5. 3 Data and sample
  6. 4 Research design
  7. 5 Test results and discussion
  8. 6 Concluding remarks
  9. Appendix
  10. References

Measures of readability

As mentioned above, we provide evidence based on two readability measures: straightforward language and conciseness (number of words and number of characters).

Straightforward Language

Straightforward language consists of Fog and Flesch measures. The Fog index (also known as the Gunning Fog Index) is one of the most widely used measures of readability. It is indicative of the level of formal education a reader of average intelligence would need to read the text once and understand that piece of writing clearly. It is calculated as follows:

  • display math(1)

Complex words are identified as words with three syllables or more.

The Flesch-Kincaid and the Flesch Reading Ease indices are popular among practitioners. Smith and Smith (1971) and Healy (1977) previously employed the Flesch reading scores. The Flesch-Kincaid index is calculated as follows:

  • display math(2)

The Flesch Reading Ease index is our second Flesch index. Similar to the other measures, the higher the score, the easier is the text to read. It is calculated as:

  • display math(3)

To create an aggregate measure of straightforward language (Readability-Words), we average these three indices. Specifically, we compute percentile ranks for each component (from 1–100) and then take the average across the three groups. Since it is more intuitive to interpret readability than “unreadability,” we multiply each component by negative one to ensure that the component is increasing in readability.

Conciseness

Following Li (2008), we measure readability using the length of the report. The two components are the number of words and the number of characters in the report (see also SEC 1998; Barker 2002; Lawrence 2012). The argument is that longer reports are more difficult to understand due to higher information processing costs.11 We create an aggregate measure of conciseness (Readability-Length) similar to what we do for straightforward language.12

Factors explaining the readability of analysts' reports

We use the following model to test whether high-ability analysts issue more readable reports:

  • display math(4)

The specification is at the analyst report level, which is issued about firm i by analyst k on day t. For this and all analyses in this study, we estimate t-statistics using robust standard errors clustered at both the analyst-firm and week levels to control for the analyst-firm correlation across time. All variables are defined in the Appendix.

To test for the determinants of report readability, we employ five proxies for analyst ability and expect each one of these proxies to be positively related to analysts' report readability. Our first proxy is the analyst's experience (Clement 1999; Mikhail, Walther, and Willis 1997, 2003). The longer the analyst works in the industry, the more systematic and idiosyncratic knowledge the analyst acquires. In addition, experience is, by definition, negatively correlated with turnover, and we expect that high-ability analysts will survive longer in the industry. Accordingly, we use the analyst's general experience (Experience) in our tests.

Our second proxy is whether the analyst is a leader as defined by the forecast timeliness measure of Cooper, Day, and Lewis (2001) (see also Mozes 2003). Forecast timeliness has been shown to be associated with market reactions and is measured as the average leader–follower ratio of the analyst's earnings forecasts issued in each year (Leader).

Third, II-ranked analysts exhibit higher ability as evidenced through their greater forecasting accuracy and stock recommendation profitability (e.g., Stickel 1990, 1992; Gleason and Lee 2003; Bonner, Hugon, and Walther 2007; Emery and Li 2009). We measure II-Rank as a binary variable that takes the value of one if the analyst is II-ranked in a given year, otherwise zero. For the tests of II-Rank we exclude brokers without at least one II-ranked ranked analyst in that year because these brokers typically do not serve a broad group of institutional investors, which is essential in order to gather enough votes to be ranked.

Fourth, we consider the intensity with which the analyst covers the firm, as measured by the number of forecast revisions per year per firm (Fcst Frequency). The more often the analyst updates the earnings forecast, the more likely the analyst is well informed about the operations of the company (e.g., Jacob, Lys, and Neale 1999).

Our last test variable measures analysts' propensity for consistent outputs. Brown and Huang (2013) document that analysts' earnings forecasts are often inconsistent with their stock recommendations when issued at the same time. They further show that consistent forecast-recommendation pairs (i.e., when both the forecast and the recommendation are above or below its existing consensus) are more informative to investors than inconsistent pairs. We test the effect of consistency for a subsample of observations for which forecasts and recommendations are updated simultaneously. However, this last measure could also proxy for “obfuscation” incentives (Li 2008), as analysts who provide inconsistent forecast-recommendation outputs might potentially conceal their inconsistency through unreadable reports.

We test for the effects of these five analyst characteristics simultaneously to pick up the incremental effect of each analyst attribute.

We control for several other analyst characteristics in our tests. We first consider the role of analyst task complexity. In particular, we test whether writing reports on firms in multiple industries affects readability (Industries Covered). Going across industries makes it more difficult to use prior experience in an industry, which, ceteris paribus, should complicate the analyst's task. We also control for the number of companies covered by the analyst (Firms Covered). This construct could have alternative interpretations. Since we also control for the number of industries followed, the number of firms covered could be viewed as a measure of the analyst's industry expertise. If this effect dominates, we expect a positive effect on readability. However, an alternative view is that it represents another dimension of complexity, which would lead to a negative relation with readability.

We measure Firm Complexity as the average of four normalized underlying variables (stock return variability, the inverse of firm age, the number of business segments, and the volatility of earnings). We expect that, ceteris paribus, the more complicated the firm's operations, the more involved and the less readable the report will be. Ramnath et al. (2008) provide an extensive survey of the literature that studies the effect of complexity on analysts' outputs such as earnings forecasts.

We include a variable to indicate whether an analyst's report also contains a forecast, recommendation, or target price revision (Analyst Rev), to control for the news of the report. We also control for a variable that indicates firm-level releases of earnings news such as earnings announcements over the five-day report window (Earn Annce). To control for the precision of forward-looking voluntary disclosures by the management, we control for the management forecast score computed in accordance with Francis, Nanda, and Olsson (2008) (Mgmt Fcst). For these news variables, we additionally include binary values that indicate whether the news is negative. Li (2008) shows that firms with bad news are more likely to make their periodic financial reports less readable.

Further, to consider the possibility that readability is potentially less important for sophisticated institutional investors, we control for firm-level institutional stock-holdings, Institutional Holdings, calculated to indicate above-median level of institutional holdings. Institutional investors are often among the primary clients of analysts and are considered more sophisticated (e.g., D'Souza, Ramesh, and Shen 2010). As such, it is conceivable that they could prefer unprocessed information in more detail, even if less readable. Our empirical tests further control for the size and the book–market ratio of the analyzed company (e.g., Li 2008). Last, we include broker, industry, and year fixed effects. Controlling for broker effects is important because analysts must typically follow their employer's policies on report formatting and content. For example, each report contains required disclosures on the broker's distribution of recommendations as well as certain disclaimers. We control for Fama-French industry membership as complexity varies across industries. To control for any systematic variation across time, all tests include year fixed effects.

Trading volume reactions to the readability of analysts' reports

We use the following regression model to test the effect of report readability on trading volume:

  • display math(5)

Volume is the logarithm of the cumulative trading volume over the three-day window centered on the analysts' report date minus the logarithm of the firm-specific median trading volume for contiguous three-day periods over the 365 days prior to the report window; hence this variable measures abnormal trading volume. As this is a short-window event study, any observed effect is less likely to be driven by omitted variables. Also, consistent with event study tests in other literatures, we argue that this approach helps in identifying a causal link. However, as explained next, we include a number of control variables. Note further that this is the second stage of a two-state least squares (2SLS) estimation. That is, Readability-Words and Readability-Length are the predicted values from the estimation of (4), and we consequently control for determinants of the readability measures when we test for volume reactions.13

Although our study centers on readability, the volume reaction is likely affected by the tone of the analyst report. As such, we test for the effect of readability after controlling for the effect of tone. For this purpose, we include the news in the analyst's report using the General Inquirer (GI) classification algorithm to determine whether the report's text has a positive or negative tone. The GI dictionary provides a list of words associated with a positive or negative tone. A word is classified as positive or negative (or neutral if it is not on the GI lists). To create a continuous measure of the tone of the report's text, we take the number of positive GI words less the number of negative GI words and scale this number by the total number of positive and negative GI words in the text. Next, we calculate the news in the report's tone, Tone Rev, which is the tone measure for the current report less the value of the tone for the previous report issued by the same analyst for the same firm. Last, to allow for asymmetric volume reactions to the news, we present separate variables for positive tone revisions (Tone Rev Pos) and negative tone revisions (Tone Rev Neg).

Because we are interested primarily in the readability of the report, we control for several other aspects of the report, including forecast (Forecast Rev Pos and Forecast Rev Neg), recommendation (Recommendation Rev Pos and Recommendation Rev Neg), and target price (Target Price Rev Pos and Target Price Rev Neg) revisions that occur during the issuance of the analyst's report. Because our dependent variable, trading volume, is unsigned, we include the absolute values of these revisions in our tests. We expect that the greater the magnitude of the revision, the greater the volume reaction; that is, positive (negative) coefficients on the positive (negative) revision variables.

We control for the timeliness of the report using the Cooper et al. (2001) leader–follower ratio calculated at the report level using the pattern of reports (Timely). We also control for the number of analyst reports issued for a particular firm over the event period (Number of Reports). We expect this variable to relate positively to economically important firm-related news released during the event window. We expect that both timeliness and a larger number of reports should be associated with a greater volume response. Next, we include earnings news (|Earning News|) and management forecast news (|Management Forecast News|). These variables control for surprises to investors and should be positively related to trading volume.

As an important firm characteristic, we control for the complexity of the firm's operations (Firm Complexity). As general firm controls, we include firms' book–market ratio and market value. Broker effects are included to control for the size of the broker's sales force and its impact on investors, which is likely to be systematically related to the volume traded per report. Last, we include Fama-French industry and year fixed effects.

5 Test results and discussion

  1. Top of page
  2. Abstract
  3. 1 Introduction
  4. 2 Background and research questions
  5. 3 Data and sample
  6. 4 Research design
  7. 5 Test results and discussion
  8. 6 Concluding remarks
  9. Appendix
  10. References

Descriptive statistics and correlations

Before reporting the results of our analyses, Table 1 provides the descriptive statistics and shows that our sample is comparable to the samples used in previous literature. Although we use aggregate readability scores in our tests, we also provide descriptive statistics for the five components of readability. The mean (median) value of our first readability component, Fog, is 18.7 (18.4), which is close to the mean (median) value of 19.3 (19.2) reported by Li (2008). This readability score suggests that analyst reports (and annual reports in Li 2008) presume a basic understanding of complex business/financial language. The mean (median) value of the Flesch Reading Ease index is 8.5 (8.7).

Table 1. Descriptive statistics
    Percentile
Variable N MeanSt. dev.25thMedian75th
Notes
  1. This table presents descriptive statistics of the variables used in our tests. Variables are defined in the Appendix.

Report Readability
Readability-Words 356,463−49.5022.56−67.33−47.67−31.67
Readability-Length 356,463−49.5028.51−74.00−49.50−25.00
Fog 356,46318.712.4216.9218.4220.08
Flesch-Kincaid 356,463−51.8513.99−57.86−49.53−42.87
Flesch Reading Ease 356,4638.482.347.098.749.98
Length Words 356,4637.880.567.527.918.26
Length Characters 356,4639.720.579.369.7410.07
Market Reaction
Volume 356,4630.470.620.050.410.82
|Ret| 356,4630.050.050.010.030.06
Analyst Ability
Experience 356,46312.877.426.0014.0019.00
Leader 356,4635.083.382.624.196.71
II-Rank 212,4160.400.490.000.001.00
Fcst Frequency 356,4638.143.156.227.729.62
Fcst-Rec Consistency 299,1910.490.240.330.500.67
Control
Tone Rev Pos 356,4630.020.030.000.000.03
Tone Rev Neg 356,463−0.020.04−0.030.000.00
Fcst Rev Pos 356,4630.000.010.000.000.00
Fcst Rev Neg 356,463−0.000.01−0.000.000.00
Recommendation Rev Pos 356,4630.030.210.000.000.00
Recommendation Rev Neg 356,463−0.030.210.000.000.00
Target Price Rev Pos 356,4630.020.080.000.000.00
Target Price Rev Neg 356,463−0.020.080.000.000.00
Timely 356,4633.787.410.291.003.41
Number of Reports 356,4633.512.901.002.005.00
|Earnings News|356,4630.000.010.000.000.00
|Mgmt Fcst News |356,4630.000.000.000.000.00
Firm Complexity 356,46348.3514.9938.2548.7558.50
Book-Market 356,4630.430.320.220.360.56
Market Value 356,4637.941.786.657.799.18
Industries Covered 356,4634.502.433.004.006.00
Firms Covered 356,46316.506.8112.0016.0020.00
Analyst Rev 356,4630.600.490.001.001.00
Earn Annce 356,4630.400.490.000.001.00
Mgmt Fcst 356,4630.390.840.000.000.00
Analyst Rev Neg 356,4630.290.460.000.001.00
Earn Annce Neg 356,4630.140.350.000.000.00
Mgmt Fcst Neg 356,4630.010.110.000.000.00
Institutional Holdings 356,4630.400.490.000.001.00

The mean (median) value of the abnormal trading volume over the three-day report window is 0.47 (0.41). This is comparable to the mean (median) value of the abnormal trading volume over the three-day quarterly earnings announcement window of 0.53 (0.48) reported by Barron et al. (2010). Regarding analyst ability characteristics, analysts have a mean 12.9 years of experience. The mean value of Leader is 5.1. For each firm, each analyst has a mean 8.1 forecast revisions per year. The mean value of II-Rank is 0.40, which implies that approximately 40 percent of reports in the subsample of brokers with an II-ranked analyst are written by II-ranked analysts.14 When analysts issue both a forecast and recommendation updates in conjunction with a report, the mean value of Fcst-Rec Consistency indicates that in almost half of the cases the analyst has a consistent forecast and recommendation (i.e., both earnings forecast and recommendation are above the consensus or both are below the consensus). We additionally note that the average analyst in our sample follows 16 to 17 companies across four to five industries.

Pearson correlations between the independent variables for the determinants and consequences tests are presented in Tables 2 and 3, respectively. Many of the variables are significantly correlated with each other, although most correlation coefficients are not large in magnitude. Some of the largest correlations are between Industries Covered and Firms Covered (0.41), Earn Annce and Analyst Rev (0.36), and between Firm Complexity and Market Value (−0.33).15

Table 2. Pearson correlations between independent variables in analyst readability (Table 4) tests
  (I)(II)(III)(IV)(V)(VI)(VII)(VIII)(IX)(X)(XI)(XII)(XIII)(XIV)(XV)(XVI)
Notes
  1. This table reports Pearson correlations between the independent variables used in the Table 5 tests. Variables are defined in the Appendix.

  2. a

    Significant at the 1 percent level (two-tailed).

Experience (I)                
Leader (II)0.07a               
II-Rank (III)0.18a0.01a              
Fcst Frequency (IV)0.02a−0.05a0.12a             
Fcst-Rec Consistency (V)0.02a−0.02a0.06a0.03a            
Industries Covered (VI)0.12a0.14a0.11a0.02a−0.02a           
Firms Covered (VII)0.25a0.05a0.14a0.09a0.01a0.41a          
Analyst Rev (VIII)0.02a0.02a0.01a0.14a−0.01a0.06a0.08a         
Earn Annce (IX)0.01a0.04a−0.02a0.00−0.02a0.04a0.05a0.36a        
Mgmt Fcst (X)0.02a0.07a0.01a−0.02a−0.000.08a−0.01a0.18a0.33a       
Analyst Rev Neg (XI)0.01a−0.00−0.01a0.09a−0.000.04a0.05a0.52a0.17a0.07a      
Earn Annce Neg (XII)0.01a0.00−0.01a0.02a−0.01a0.02a0.03a0.16a0.49a0.09a0.23a     
Mgmt Fcst Neg (XIII)−0.000.01a−0.000.01a0.01a0.01a−0.000.05a−0.08a0.17a0.10a−0.04a    
Firm Complexity (XIV)−0.13a−0.03a−0.03a−0.02a−0.000.06a−0.05a0.03a0.00a−0.08a0.02a0.01a0.01a   
Book-Market (XV)0.05a−0.06a0.000.05a−0.000.01a0.06a0.04a0.03a−0.05a0.05a0.06a−0.01a0.10a  
Market Value (XVI)0.14a0.02a0.15a0.16a0.02a−0.08a−0.01a−0.07a−0.06a0.06a−0.06a−0.07a−0.00−0.33a−0.22a 
Institutional Holdings (XVII)−0.01a0.04a0.03a0.03a0.01a0.06a−0.03a0.01a0.02a0.04a−0.00−0.01a0.02a0.06a−0.01a0.04a
Table 3. Pearson correlations among independent variables in trading volume (Table 5) tests
  (I)(II)(III)(IV)(V)(VI)(VII)(VIII)(IX)(X)(XI)(XII)(XIII)(XIV)(XV)(XVI)
Notes
  1. This table reports Pearson correlations between the independent variables used in the Table 4 tests. Variables are defined in the Appendix.

  2. a

    Significant at the 1 percent level (two-tailed).

Readable-Words (I)                
Readable-Length (II)0.13a               
Tone Rev Pos (III)0.01a0.05a              
Tone Rev Neg (IV)−0.05a−0.13a0.34a             
Forecast Rev Pos (V)0.04a0.01a0.01a0.00            
Forecast Rev Neg (VI)−0.05a−0.02a−0.00−0.000.06a           
Recommendation Rev Pos (VII)0.02a0.000.01a−0.01a0.06a0.00          
Recommendation Rev Neg (VIII)−0.02a−0.03a−0.000.02a0.01a0.07a0.02a         
Target Price Rev Pos (IX)0.02a−0.04a0.01a−0.000.16a−0.000.15a0.01a        
Target Price Rev Neg (X)−0.01a0.02a−0.000.01a0.000.18a−0.02a0.13a0.05a       
Timely (XI)−0.03a0.01a−0.02a−0.01a−0.00−0.01a−0.01a0.000.00−0.00      
Number of Reports (XII)−0.01a−0.11a0.03a0.01a−0.000.01a−0.03a0.02a0.01a−0.01a0.03a     
|Earnings News|(XIII)0.04a0.01a0.000.02a0.15a−0.20a−0.01a0.000.05a−0.05a0.06a0.06a    
|Mgmt Fcst News |(XIV)0.01a0.02a−0.01a−0.02a0.01a−0.02a−0.00−0.01a0.01a−0.02a0.01a0.03a−0.02a   
Firm Complexity (XV)−0.01a−0.01a−0.01a0.01a0.08a−0.09a0.01a−0.02a0.06a−0.07a−0.01a−0.04a0.14a0.01a  
Book-Market (XVI)0.02a0.05a−0.01a0.01a0.07a−0.14a0.01a−0.01a0.05a−0.04a0.03a−0.08a0.20a0.02a0.10a 
Market Value (XVII)−0.13a−0.20a0.01a−0.00−0.10a0.12a−0.03a0.04a−0.08a0.07a−0.04a0.26a−0.15a−0.01a−0.33a−0.22a
Table 4. Tests of factors explaining analysts' report readability
Explanatory variablePred. signDep. var. = Readability-WordsDep. var. = Readability-Length
  123456
Notes
  1. This table reports an analysis of a model that explains variations in the readability of analysts' reports. It summarizes the results of regressing report readability on analyst and firm characteristics. Broker, industry and year fixed effects are included for each model but not tabulated. We estimate t-statistics using robust standard errors clustered at both analyst-firm and week levels. Coefficient t-statistics are in parentheses.

  2. *, **, and *** represent significance levels of 10 percent, 5 percent, and 1 percent (two-sided), respectively.

  3. Variables are defined in the Appendix.

Experience +0.04**0.030.04**0.13***0.24***0.11***
 (2.50)(1.55)(2.12)(8.17)(12.25)(6.54)
Leader +0.09***−0.010.09**0.33***0.24***0.36***
 (2.85)(−0.32)(2.54)(9.98)(6.51)(9.49)
II-Rank + 0.59**  −1.47*** 
  (2.00)  (−5.04) 
Fcst Frequency +0.35***0.30***0.37***−0.000.070.04
 (9.78)(7.33)(9.13)(−0.06)(1.60)(0.92)
Fcst-Rec Consistency +  1.10***  1.10***
   (3.03)  (2.72)
Industries Covered −0.28***−0.23***−0.23***−0.00−0.24***0.03
 (−4.57)(−2.96)(−3.45)(−0.06)(−3.48)(0.47)
Firms Covered ?0.14***0.06***0.15***0.07***0.010.12***
 (7.55)(2.77)(7.22)(2.84)(0.24)(4.67)
Analyst Rev ?4.59***4.96***4.70***−2.71***−4.49***−2.54***
 (27.75)(22.78)(26.87)(−12.33)(−15.82)(−10.83)
Earn Annce ?3.98***3.93***3.90***−1.24***−1.28***−1.28***
 (25.15)(19.38)(23.58)(−3.34)(−2.98)(−3.43)
Mgmt Fcst ?0.39***0.30***0.39***0.46***0.41***0.60***
 (4.72)(2.98)(4.35)(4.69)(3.86)(5.88)
Analyst Rev Neg −0.07−0.16−0.040.60***0.150.63***
 (−0.63)(−1.09)(−0.34)(4.16)(0.80)(4.35)
Earn Annce Neg −0.66***−0.65***−0.61***0.91***0.90***0.98***
 (−4.48)(−3.51)(−4.12)(5.95)(4.97)(6.08)
Mgmt Fcst Neg 0.600.470.425.50***5.02***5.27***
 (1.60)(1.02)(1.08)(10.76)(7.80)(9.76)
Firm Complexity −0.04***−0.03***−0.04***−0.01*0.02**−0.01
 (−5.18)(−3.36)(−4.52)(−1.94)(1.68)(−1.14)
Book-Market ?−0.46−0.10−0.75**1.64***0.521.73***
 (−1.49)(−0.26)(−2.26)(5.18)(1.38)(5.19)
Market Value −0.20**−0.13−0.18**−0.94***−1.06***−0.90***
 (−2.55)(−1.29)(−2.07)(−11.31)(−11.02)(−10.31)
Institutional Holdings ?−0.05−0.33−0.08−0.25−0.11−0.24
 (−0.26)(−1.52)(−0.39)(−1.39)(−0.51)(−1.26)
Broker fixed effects YesYesYesYesYesYes
Industry fixed effects YesYesYesYesYesYes
Year fixed effects YesYesYesYesYesYes
No. of obs. 356,463212,416299,191356,463212,416299,191
Adj. R2 (%) 35.7537.9435.9538.3832.4839.58
Table 5. Tests of capital market consequences of analysts' report readability
Explanatory variablePred. signDep. var. = Volume
Notes
  1. This table reports an analysis of the relation between trading volume and the readability of analysts' reports. It summarizes the results of regressing abnormal trading volume on report readability, controlling for other report and firm characteristics. Broker, industry and year fixed effects are included for each model but not tabulated. We estimate t-statistics using robust standard errors clustered at both analyst-firm and week levels. To ease interpretation we multiply coefficients by 100. Coefficient t-statistics are in parentheses.

  2. * and ** represent significance levels of 10 percent and 1 percent (two-sided), respectively.

  3. Variables are defined in the Appendix.

Readability-Words +2.22**
 (27.82)
Readability-Length +0.44**
 (4.30)
Tone Rev Pos +0.11**
 (2.89)
Tone Rev Neg −0.19**
 (−5.42)
Fcst Rev Pos +0.48*
 (1.77)
Fcst Rev Neg −0.16
 (−0.87)
Recommendation Rev Pos +0.09**
 (14.36)
Recommendation Rev Neg −0.20**
 (−24.85)
Target Price Rev Pos +0.51**
 (19.29)
Target Price Rev Neg −0.63**
 (−18.98)
Timely +0.00
 (0.09)
Number of Reports +0.04**
 (38.10)
|Earnings News|+3.67**
 (6.01)
|Mgmt Fcst News |+15.61**
 (11.04)
Firm Complexity ?0.00**
 (7.90)
Book-Market ?−0.09**
 (−10.40)
Market Value −0.05**
 (−19.99)
Broker fixed effects Yes
Industry fixed effects Yes
Year fixed effects Yes
No. of obs. 356,463
Adj. R2 (%) 14.19

Determinants of report readability

We first use this sample to examine the factors that explain variations in analysts' report readability by estimating (4). Column 1 of Table 4 presents the results of the multivariate regressions for Readability-Words using the full sample. The results for three of our high-ability-analyst proxies employed in the unrestricted sample provide support for the idea that high-ability analysts issue more readable reports. Specifically, the estimated coefficients on Experience, Leader, and Fcst Frequency are positive and statistically significant. Results are generally consistent when using Readability-Length as the dependent variable in column 4, with the exception of Fcst Frequency, which is statistically insignificant. These findings are consistent with individuals who have worked longer as an analyst and who publish more timely earnings forecasts writing more readable reports.

Column 2 reports that II-ranked analysts write more readable reports. However, interestingly we observe in column 5 that they also produce longer reports than do other analysts. In columns 3 and 6, we test for the effect on readability of analysts who provide consistent revisions for their earnings forecasts and stock recommendation. The results for the first three test variables in column 1 are consistent with those found in this subsample test. More importantly, and in line with the view of consistency as a dimension of analyst ability (e.g., Brown and Huang 2012), Fcst-Rec Consistency is positively and significantly associated with readability. An alternative explanation could be that more consistent analysts have less to cover up by issuing unreadable reports.16

Given the large number of control variables included and in particular the extensive fixed effects structure employed in our models, it is perhaps not surprising that if economic significance is defined as the incremental explanatory power of the model from adding the test variables, the economic importance is modest. To be specific, when Readability-Length (Readability-Words) is the dependent variable, the increase in adjusted R2 is almost two percentage points (less than one percentage point).

With respect to the control variables, they mostly load as predicted. Analysts issue less readable reports if they operate in a more complex work environment in that they cover multiple industries (Industries Covered). As discussed above, we have no signed prediction for Firms Covered but the estimated coefficients are mostly positive and significant. Moreover, analysts' reports are more readable for firms whose operations are less complex (Firm Complexity). This is an intuitive finding and provides further validity to the measurement of our dependent variable. In the case of the news variables, reports around earnings announcements use more straightforward language (columns 1–3) but tend to be less concise (see columns 4–6).

The results further indicate that reports tend to be more readable and concise when a report is issued around more precise management forecasts and more straightforward but less concise when there is an analyst revision (i.e., a forecast, recommendation, or target price revision). As expected, there is also evidence that when earnings announcements are negative, reports are less straightforward; in addition the reports are less concise. Further, we find that readability is decreasing in firm size. With regard to Institutional Holdings, we find no significant relation between firm-level institutional stockholdings and readability. This could reflect that institutional stockholders do not specifically demand or value more readable reports but also do not have any incentives to seek less readable reports. This ambivalence manifests in an insignificant statistical relation.

Taken together, the evidence suggests that variations in the readability of analyst reports are positively related to proxies for high-ability analysts and hence supports the idea that high-ability analysts write more readable reports.

Consequences of report readability

We continue by testing our second hypothesis, as described in (5). Table 5 summarizes the results of our estimation of (5). The positive coefficient on Readability-Words, 2.22, significant at the one percent level using a two-tailed test, provides evidence that report readability in terms of straightforward language is positively related to the short-window abnormal volume reaction. We find consistent results for Readability-Length, which has a coefficient estimate of 0.44 and is significant at the one percent level. We find that the adjusted R2 increases by two percentage points when Readability-Words is added but by less than one percentage point when Readability-Length is added. These results suggest that Readability-Words is economically important but that Readability-Length only has marginal economic significance (albeit still strongly statistically significant).17

These findings are consistent with analysts generating trading volume by providing more precise information in their reports to investors that causes them to initiate trades (e.g., Karpoff 1987; Kim and Verrecchia 1991). It is not consistent with the null hypothesis that analysts generate trade by issuing less readable reports with the aim of creating divergence of opinions among investors.

As expected, volume is increasing in the magnitude of the revisions to the tone of the reports, forecasts, recommendations, and target prices, as indicated by the positive (negative) coefficients on the positive (negative) revisions.18 In addition, reports that are accompanied on the same day by other reports are positively associated with volume. As predicted, volume is also increasing in the magnitude of earnings news, management earnings forecast news, and the complexity of the firm. Firms with higher book–market ratios and that are larger tend to have smaller volume reactions to analyst reports.

Additional analyses

Speed of communications

An advantage of the analyst report setting over annual reports is that analyst reports are likely less heavily edited before being released and hence lend themselves more to tests of lexical analysis that rely on the communications being relatively “spontaneous.” As a further test of the importance of the speed of communication, we create an indicator variable equal to one for analyst reports issued within seven days of an earnings announcement, and zero otherwise (SevenDays). These analyst reports are less likely to be vetted by people other than the analyst and his team. We then interact this indicator variable with words and length, respectively. Panel A of Table 6 shows that the interaction between Readability-Words and SevenDays is positive and statistically significant at the one percent level (whereas the interaction with Readability-Length is not significant). Thus there is some indication that the economic importance of readability is higher for analyst reports that are more likely to represent analysts' unconscious behaviors and thought processes.

Table 6. Additional analyses and robustness tests
Panel A: Speed of communications
Explanatory variableDep. var. = Volume
Readability-Words × SevenDays 0.08***
(3.07)
Readability-Length × SevenDays −0.00
(−0.09)
Panel B: Interaction with stock recommendation revisions
Explanatory variableDep. var. = Volume
Readability-Words × Recommendation Rev Pos 0.13***
(2.89)
Readability-Words × Recommendation Rev Neg −0.12**
(−2.25)
Readability-Length × Recommendation Rev Pos −0.04
(−1.07)
Readability-Length × Recommendation Rev Neg 0.06
(1.16)
Panel C: Interactions between readability and tone
Explanatory variableDep. var. = Volume
Readability-Words × Tone 0.40**
(2.23)
Readability-Length × Tone 0.06
(0.43)
Panel D: Role of annual report readability-explaining analysts' report readability
Explanatory variableDep. var. = Readability-Words Dep. Var. = Readability-Length
Annual Report Readability-Words 0.02*** 0.00
(4.89)(0.15)
Annual Report Readability-Length 0.02*** −0.00
(4.30)(−0.25)
Panel E: Role of annual report readability-capital market consequences
Explanatory variableDep. var. = Volume
Readability-Words × Annual Report Readability-Words 0.00
(0.09)
Readability-Length × Annual Report Readability-Length −0.08**
(−2.07)
Panel F: Investment banking ties
Explanatory variableDep. var. = Readability-Words Dep. var. = Readability-Length
IB 0.63−19.61***
(0.89)(−37.48)
Panel G: Longer-term economic consequences
Explanatory variableDep. var. = Stock Price Volatility
Readability-Words −2.53*
(−1.78)
Readability-Length −4.94***
(−3.75)
Panel H: Using abnormal returns instead of trading volume
Explanatory variableDep. var. = |Ret|
Readability-Words 0.09***
(16.81)
Readability-Length 0.02**
(2.53)
Panel I: Changes specification
Explanatory variableDep. var. = Volume
Notes
  1. This table reports various additional analyses and robustness tests. For brevity we only tabulate coefficients on the variables of interest. All control variables, including broker, industry, and year fixed effects, are included for each model but not tabulated. We estimate t-statistics using robust standard errors clustered at both analyst-firm and week levels. To ease interpretation we multiply coefficients by 100. Coefficient t-statistics are in parentheses.

  2. *, **, and *** represent significance levels of 10 percent, 5 percent, and 1 percent (two-sided), respectively.

  3. Variables are defined in the Appendix.

Changes in Readability-Words 0.90***
(19.18)
Changes in Readability-Length 0.19**
(2.56)
Interaction with stock recommendation revisions

We examine how the readability of a report affects the market's reaction to recommendation update. We interact the two readability measures separately with positive and negative recommendation revisions (i.e., upgrades and downgrades). In panel B the interactions between Readability-Words and both upgrades and downgrades are statistically significant (whereas the interactions with Readability-Length are not significant), consistent with readability strengthening the market reaction to analysts' recommendation revisions.

Interactions between readability and tone

We next explore whether the economic relevance of readability varies with the type of news conveyed by the analyst report as measured by the report's tone. We do not have a directional prediction for this test. Panel C shows that the interaction between Readability-Words and Tone is positive and statistically significant, suggesting that tone and readability reinforce each other.

Role of annual report readability

We consider the additional role of annual report readability in our setting. We use annual reports issued within 365 days of the analyst report and obtain annual report readability scores from Feng Li's website.19 We first test whether the readability of analyst reports is influenced by the readability of the annual reports and present the results in panel D. The coefficients on both of Li's readability measures are positively and significantly (not significantly) associated with Readability-Words (Readability-Length). Inferences for the analyst ability variables are not affected by the inclusion of annual report readability (untabulated). We next investigate whether the benefit of readability of analyst reports varies with the opaqueness of annual reports. Panel E reports that the interaction between the two Fog-based variables is not significant; but the interaction between the two length-based variables is negative and significant, providing some limited evidence for a substitutive effect between analyst report and annual report readability on trading volume.

Investment banking ties

We consider broker fixed effects a strong control and particularly important in our setting as report formats typically vary by brokerage house. However, to explore the role of investment banking affiliations, we replace these fixed effects with an indicator variable (IB) equal to one if the client firm has an investment banking affiliation with the particular broker firm, zero otherwise.20 We observe an insignificant effect of IB on Readability-Words but a negative and significant effect on Readability-Length in panel F. The latter could potentially be viewed as consistent with Li's “obfuscation theory.”

Longer-term economic consequences

Our emphasis on short-window market reactions has the advantage of providing a strong control for confounding events. We further explore longer-term economic consequences in panel G. We define longer-term consequences as one-year-ahead stock price volatility (scaled by the mean return over the same period). We find negative and significant estimated coefficients on both Readability-Words and Readability-Length in these regressions.

Additional robustness tests

Using abnormal returns instead of trading volume

We measure abnormal returns as the logarithm of one plus the absolute value of sized-adjusted returns over the three-day window centered on report dates. Consistent with our volume tests, in panel H, we find that the coefficients on both readability measures are positively and significantly (at the 1 percent level using a two-tailed test) associated with abnormal returns.

Changes specification, firm fixed effects, and analyst fixed effects

Although we use short-window tests and include numerous controls, we report results of three important sensitivity tests. First and most importantly, we use a changes specification. Changes tests are useful for controlling for omitted variables that do not change over time. Specifically, we first estimate the first-stage determinants regressions and calculate predicted values of Readability-Words and Readability-Length. In the second stage we take first-order differences of all (time-varying) variables. Most importantly, for test variables we compute the change as the current minus the lagged value where the latter is the readability of the report that the same analyst wrote about the same firm, issued immediately prior to the current report. We find that the estimated coefficients on the change in the test variables are positively and statistically significant at the one percent level (see panel I). This finding helps rule out alternative explanations.

Second, in untabulated analyses, as an additional control for unobservable firm characteristics, we reestimate the test including firm fixed effects instead of industry fixed effects. Finally, we replace broker fixed effects with analyst fixed effects. No inferences are affected with these alternative fixed effects specifications.

Excluding “Morning notes”

We exclude from our sample reports that are shorter than 300 words. This exclusion implies that our tests are not picking up the effects of the shortest notes. Our empirical tests also control for the number of analyst reports issued for a particular firm over the five-day report window. As an additional untabulated test, we exclude all “morning meeting notes” (which tend to be much shorter than other detailed investment reports). Results (untabulated) are similar to those reported and no inferences are affected.

Excluding observations during earnings announcement windows

Our empirical analyses control for the absolute value of earnings news (|Earnings News|), the existence of an earnings announcement (Earn Annce), and whether it is negative news (Earn Annce Neg). In addition, we control for several analyst variables that tend to cluster around earnings announcements. However, as an alternative to explicitly controlling for earnings announcement effects through control variables, in untabulated analyses, we reestimate our tests after excluding all observations that occur during the earnings announcement windows. No conclusions are affected in this smaller sample for the trading volume tests. For the determinants tests, the results for Experience, Frequency, and Fcst-Rec Consistency are similar, and no inferences are impacted using either Readability-Words or Readability-Length as dependent variables. For Leader and II-Rank, the results are similar to those tabulated when Readability-Length is the dependent variable. However, they are no longer statistically significant when tested against Readability-Words.

6 Concluding remarks

  1. Top of page
  2. Abstract
  3. 1 Introduction
  4. 2 Background and research questions
  5. 3 Data and sample
  6. 4 Research design
  7. 5 Test results and discussion
  8. 6 Concluding remarks
  9. Appendix
  10. References

Although a large body of research examines analysts' earnings forecasts or stock recommendations, research on the written disclosures provided by analysts is limited. This study examines the readability of analysts' reports. Analysts' reports likely fit the setting that Bloomfield (2008) calls for: more spontaneous and timely communications than annual reports. We use a large database of the text of analysts' reports from 2002 to 2009. Our evidence indicates “high-ability” analysts, broadly defined to include analysts with more experience, more timely earnings forecasts, more frequent forecast revisions, analysts who are ranked by Institutional Investor magazine, and who are consistent in their forecast-recommendation revisions issue reports that are more readable.

We further show that trading volume reactions are increasing in the readability of analysts' reports, consistent with the idea that readability affects investors' decisions. This finding is consistent with theoretical models proposing that more precise, and hence more informative, signals cause investors to initiate trades. These results support the notion that the readability of their reports is important for analysts and for the users of their reports. However, as discussed above, although the results are strongly statistically significant, the increases in the adjusted R2s from adding our test variables are modest.

Appendix

  1. Top of page
  2. Abstract
  3. 1 Introduction
  4. 2 Background and research questions
  5. 3 Data and sample
  6. 4 Research design
  7. 5 Test results and discussion
  8. 6 Concluding remarks
  9. Appendix
  10. References

Variable definitions

VariableDefinition
Report Readability
Readability-Words To form this aggregate readability measure, we rank each component into percentiles—Fog, Flesch-Kincaid, and Flesch Reading Ease—from 1 to 100 and then take the average across the three components. We multiply Fog and Flesch-Kincaid by negative one to ensure that all components are increasing in readability
Readability-Length To form this aggregate readability measure, we rank each component into percentiles—Length Words, Length Characters—from 1 to 100 and then take the average across the two components
Fog The Gunning-Fog Unreadability Index, calculated as (words per sentence + percent of complex words) × 0.4
Flesch-Kincaid The Flesch-Kincaid Grade Level index, calculated as (11.8 ×  syllables per word) + (0.39 ×  words per sentence) – 15.59
Flesch Reading Ease −1 ×  (Flesch Reading Ease score). The Flesch Reading Ease score is calculated as 206.835 – (1.015 ×  words per sentence) – (84.6 ×  syllables per word)
Length Words Logarithm of the number of words in the report
Length Characters Logarithm of the number of characters in the report
Market Reaction
Volume The excess trading volume, defined as the logarithm of the cumulative trading volume over the three-day report window (days −1 to +1 relative the analysts' report date) minus the logarithm of the firm-specific median trading volume for contiguous three-day periods over the 365 days prior to the report window
|Ret| Absolute value of abnormal stock returns, which is the logarithm of 1 +  the absolute value of sized-adjusted returns over the three-day window centered on report dates
Analyst Ability
Experience The analyst's years of experience forecasting earnings in I/B/E/S
Leader The average leader–follower ratio of analyst j's earnings forecasts issued in each year. To form this measure, we extract all earnings forecasts for each analyst in our sample. We compare the number of days between it and the forecasts by any other analysts that precede and follow it. Let LeadingDaysijt (FollowingDaysijt) equal the total number of days between analyst j's forecast of firm i issued on day t and the two most recent preceding (following) forecasts of firm i by any other analysts. Leader–follower Ratio is set equal to LeadingDaysijt /FollowingDaysijt. We then calculate an average leader–follower ratio for each analyst in each year
II-Rank The analyst's Institutional Investor magazine rank in each year. It is set to 1 if analyst j is ranked by Institutional Investor in year t, 0 otherwise
Fcst Frequency The total number of firm forecast revisions issued by the analyst in each year
Fcst-Rec Consistency The number of consistent forecast-recommendation pairs issued by analyst j in each year. A forecast-recommendation pair is classified as consistent when both the forecast and the recommendation are above their existing consensuses or when both are below their existing consensuses
Control
Tone Rev Pos This variable is set to equal the news in the analyst's report when the news is positive, and 0 when the news is negative. The news in the analyst's report is measured as follows. The General Inquirer (GI) classification algorithm is used to determine whether the report's text has a positive or negative tone. We take the number of positive GI words less the number of negative GI words, scaled by the total number of positive and negative GI words in the text. The news in the report's tone is the tone measure for the current report less the value of the tone of the previous report issued by the same analyst for the same firm
Tone Rev Neg This variable is set to equal the news in the analyst's report when the news is negative, and 0 when the news is positive
Forecast Rev Pos This variable is set to equal analysts' forecast revisions over the five-day report window when the forecast revisions are positive, and 0 when the forecast revisions are negative. Forecast revision is defined as analyst j's earnings forecast issued in the five-day report window minus analysts j's prior earnings forecast, scaled by the stock price at the beginning of year t. If analyst j does not make any forecasts in the five-day report window, this variable is set to 0
Forecast Rev Neg This variable is set to equal analysts' forecast revisions over the five-day report window when the forecast revisions are negative, and 0 when the forecast revisions are positive
Recommendation Rev Pos This variable is set to equal analysts' recommendation revisions over the five-day report window when the recommendation revisions are positive, and 0 when the recommendation revisions are negative Forecast recommendation is defined as analyst j's stock recommendation issued in the five-day report window minus analyst's j's prior recommendation. If analyst j does not make any recommendations in the five-day report window, this variable is set to 0
Recommendation Rev Neg This variable is set to equal analysts' recommendation revisions over the five-day report window when the recommendation revisions are negative, and 0 when the recommendation revisions are positive
Target Price Rev Pos This variable is set to equal analysts' price target forecast revisions over the five-day report window when the price target forecast revisions are positive, and 0 when the price target forecast revisions are negative. Price target forecast revision is defined as analyst j's price target forecast issued in the five-day report window minus analysts j's prior price target forecast, scaled by the stock price at the beginning of year t. If analyst j does not make any price target forecasts in the five-day report window, this variable is set to 0
Target Price Rev Neg This variable is set to equal analysts' price target forecast revisions over the five-day report window when the price target forecast revisions are negative, and 0 when the price target forecast revisions are positive
Timely The leader–follower ratio at the report level. For each analyst report in our sample we compare the number of days between it and the reports by any other analysts that precede and follow it. Let LeadingDaysit (FollowingDaysit) equal the total number of days between a report of firm i issued on day t and the two most recent preceding (following) reports of firm i. Leader is set equal to LeadingDaysit /FollowingDaysit
Number of Reports The number of analysts' reports issued for firm i over the five-day report window
|Earnings News|The absolute value of earnings news over the five-day report window. Earnings news is defined as firm i's earnings announced in the five-day report window minus analysts' I/B/E/S consensus in the prior 90 days, scaled by the stock price at the beginning of year t. If firm i does not make any earnings announcement in the five-day report window, this variable is set to 0
|Mgmt Fcst News|The absolute value of management forecast news over the five-day report window. Management forecast news is defined as firm i's management forecast issued in the five-day report window minus analysts' consensus in the prior 90 days, scaled by the stock price at the beginning of year t. If firm i does not make any management forecasts in the five-day report window, this variable is set to 0
Firm Complexity The complexity of the firm. It is based on four underlying variables: stock return variability, the inverse of firm age, the number of business segments, and earnings variability. Specifically, we first normalize the three variables and then take the average of each as our summary complexity measure
Book-Market The book-to-market ratio at the beginning of the fiscal year
Market Value The logarithm of market value at the beginning of the fiscal year
Industries Covered The number of two-digit Standard Industrial Classification (SIC) industries that the analyst follows in each year
Firms Covered The number of companies the analyst follows in each year
Analyst Rev The indicator variable for analyst j's revisions issued over the five-day report window. It is set to 1 if analyst j issues at least one earnings forecast/stock recommendation/price target revision for firm i in the report window, and 0 otherwise
Earn Annce The indicator variable for earning announcements over the five-day report window. It is set to 1 if firm i issues an earnings announcements in the report window, and 0 otherwise
Mgmt Fcst The value of management forecast score (Francis et al. 2008) over the five-day report window. Francis et al. (2008) measure management forecasts by first assigning a score based on forecast occurrence and forecast specificity (0 for no forecast, 1 for qualitative forecast, 2 for range forecast, and 3 for point forecast) and then sum over all forecasts (i.e., frequency count). If firm i does not make any management forecasts in the five-day report window, this variable is set to 0
Analyst Rev Neg The indicator variable for negative revisions issued by analyst j for firm i over the five-day report window. It is set to 1 if the number of negative revisions issued by analyst j is greater than the number of positive revisions in the report window, and 0 otherwise
Earn Annce Neg The indicator variable for negative earnings announcements over the five-day report window. It is set to 1 if firm i makes negative earnings announcements in the report window, and 0 otherwise
Mgmt Fcst Neg The indicator variable negative management forecasts over the five-day report window. It is set to 1 if firm i issues negative management forecasts in the report window, and 0 otherwise
Institutional Holdings The indicator variable for high institutional holdings. It is set to 1 if firm i has above-median institutional stock holdings at the beginning of the fiscal year, and 0 otherwise
Notes
  1. 1

    For example, in a recent survey reported in Institutional Investor magazine (October 2010), investors were asked to rate the importance they place on a dozen sell-side equity analyst research attributes. The highest ranked attribute was industry knowledge. One way that analysts impart their industry knowledge is through their written reports. In addition, written reports per se were ranked fifth. For comparison, earnings estimates ranked last of the twelve attributes

  2. 2

    In additional analyses, we find some evidence that the speed of communications matters. That is, the volume reaction to readability is greater for analyst reports issued soon after earnings announcements than for other analyst reports. This suggests that the economic importance of readability is higher for analyst reports that are less edited and rehearsed and thus more likely to represent analysts' unconscious behaviors and thought processes. We further find results consistent with readability strengthening the market reaction to analysts' recommendation revisions and with readability and the tone of the report reinforcing each other. Next we find that the readability of analyst reports and annual reports are positively correlated but that controlling for annual report readability does not change any inferences. We also find corroborating results using the absolute value of abnormal returns as an alternative dependent variable.

  3. 3

    A selection of other relevant studies includes Bryan (1997), Tetlock (2007), Tetlock, Saar-Tsechansky, and Macskassy (2008), Feldman et al. (2010), Rogers, Van Buskirk, and Zechman (2011), Lawrence (2012), and Bova, Dou, and Hope (2013).

  4. 4

    Our inferences contrast somewhat with Kothari et al. (2009) because they find that the text in equity reports has an insignificant impact on the cost of capital and conclude that written disclosures by equity analysts are not considered as very relevant by equity markets.

  5. 5

    This literature includes: Beaver (1968); Bamber (1986); Holthausen and Verrecchia (1988, 1990); Kim and Verrecchia (1991, 1994, 1997); Abarbanell, Lanen, and Verrecchia (1995); Bamber and Cheon (1995); Verrecchia (2001); Cready and Hurtt (2002); Hope, Thomas, and Winterbotham (2008); and Bamber, Barron, and Stevens (2011).

  6. 6

    Although not directly related to analyst reports, the Securities and Exchange Commission (SEC) believes that the readability of financial documents is important. In 2007, the SEC sought feedback from individual investors pertaining to their financial disclosure needs (Lawrence 2012). The main conclusion from this survey was that investors want “plain language”—in other words, more readable financial reports. Also, investors do not want excessive information (e.g., Li 2013).

  7. 7

    Furthermore, the arguments leading to a positive relation between readability and trading volume may be less applicable to institutional investors, who are often among the primary clients of analysts. These investors are more sophisticated and could prefer detailed unprocessed information, even if such information is less readable (e.g., Mikhail, Walther, and Willis 2007). Note, however, that we control for institutional ownership in our empirical analyses.

  8. 8

    Additional support for the “no relation” hypothesis comes from the fact that Microsoft Word has included menu-driven tools for improving readability since the early 1980s. In other words, ordinary writers, including analysts, can easily make their writing more readable. More generally, if analysts' reports undergo extensive editing, they would not satisfy the “spontaneous communication” argument that underlies the theory of lexical analysis (Bloomfield 2008).

  9. 9

    Kim and Verrecchia (1991, 302) define their construct of ambiguity as follows: “[I]t is assumed that traders are diversely informed and differ in the precision of their private prior information; they therefore respond differently to the announcement, and this leads to positive volume.”

  10. 10

    There is some evidence that quantification (through numbers and/or tables) may enhance persuasiveness of reports (e.g., Kadous, Koonce, and Towry 2005). However, it is difficult to evaluate the clarity of tables and numbers; thus we leave this aspect for future research. Similarly, although a “picture is worth a thousand words,” we do not analyze pictures and figures in analysts' reports, although there is prior research analyzing graphics in annual reports (Preston and Young 2000).

  11. 11

    The length of the report may appear less intuitive at first glance than the Fog and Flesch scores. However, the underlying idea is that, ceteris paribus, longer documents are more difficult to read and process for users of the reports. An obvious alternative explanation for a lengthy document is greater complexity; hence our multivariate tests control for several factors related to both task and firm complexity. In other words, this readability measure can be considered as the document length beyond what is explained by normal factors (i.e., “excessive” length).

  12. 12

    Untabulated analysis shows that all five components of readability are positively and significantly correlated at the 5 percent level. In particular, the correlation between the two Flesch-based measures is 0.84, which indicates these two components likely capture the similar constructs. Similarly, the correlation between the two length-based components is 0.96, and hence we expect both components to capture the length notion. Still, the components have correlations below one, suggesting that they capture somewhat different dimensions of readability. As an alternative to averaging the measures, we have employed factor analysis, which yields consistent results.

  13. 13

    In sensitivity analyses, we have directly controlled for all of these determinants in a single-stage ordinary least squares (OLS) regression. Our inferences are unaltered.

  14. 14

    As per Table 4, the sample size for our II-rank tests consists of 212,416 observations. Of those observations, 84,966 (0.4 × 212,416) are written by II-ranked analysts. Those 84,966 reports represent 23.8 percent of our full sample of 356,463 reports.

  15. 15

    Unreported variance inflation factors do not indicate the presence of serious multicollinearity in either the economic consequences or the determinants regressions. For example, the mean VIF for the Table 4 readability (Table 5 volume) regressions is 1.19 (1.08).

  16. 16

    Alternatively, consistent analysts extrapolate from forecasts to recommendations (or vice versa) using only one model, whereas inconsistent analysts may use separate models for the two outputs, which requires more discussion.

  17. 17

    Another common way to assess economic significance is to consider the interquartile ranges of the test and dependent variables. Using the actual (unpredicted) readability scores, we find that when Readability-Words [Readability-Length] moves from the 25th to the 75th percentile of its distribution, the corresponding increase in abnormal trading volume is 7 percent [5 percent] (using the median value of Volume in our sample).

  18. 18

    As described in the Appendix, the revision variables, while partitioned by the sign of the news (positive or negative), retain the original news sign. In other words, more negative values imply greater negative revisions.

  19. 19
  20. 20

    We thank Yuyan Guan, Hai Lu, and Franco Wong for providing these data.

References

  1. Top of page
  2. Abstract
  3. 1 Introduction
  4. 2 Background and research questions
  5. 3 Data and sample
  6. 4 Research design
  7. 5 Test results and discussion
  8. 6 Concluding remarks
  9. Appendix
  10. References
  • Abarbanell, J. S., W. N. Lanen, and R. E. Verrecchia. 1995. Analysts' forecasts as proxies for investor beliefs in empirical research. Journal of Accounting and Economics 20 (1): 3160.
  • Asquith, P., M. B. Mikhail, and A. S. Au. 2005. Information content of equity analyst reports. Journal of Financial Economics 75 (2): 24582.
  • Bamber, L. S. 1986. The information content of annual earnings releases: A trading volume approach. Journal of Accounting Research 24 (1): 4056.
  • Bamber, L. S., O. E. Barron, and D. E. Stevens. 2011. Trading volume around earnings announcements and other financial reports: Theory, research design, empirical evidence, and directions for future research. Contemporary Accounting Research 28 (2): 43171.
  • Bamber, L. S., and Y. Cheon. 1995. Differential price and volume reactions to accounting earnings announcements. The Accounting Review 70 (3): 41741.
  • Barker, R. 2002. A three-point plan for SEC reform. Business Week Online, http://www.businessweek.com/stories/2002-06-02/a-three-point-plan-for-sec-reform
  • Barron, O. E., D. A. Byard, and Y. Yu. 2010. Earnings announcement disclosures that spur differences in interpretation. Working paper, Penn State University, Baruch College, and University of Texas at Austin.
  • Barron, O. E., and J. M. Karpoff. 2004. Information precision, transaction costs, and trading volume. Journal of Banking and Finance 28 (6): 120723.
  • Beaver, W. 1968. The information content of annual earnings announcements. Journal of Accounting Research 6 (1): 6792.
  • Bloomfield, R. 2008. Discussion of “Annual report readability, current earnings, and earnings persistence.” Journal of Accounting and Economics 45 (2–3): 24852.
  • Bonner, S., A. Hugon, and B. R. Walther. 2007. Investor reaction to celebrity analysts: The case of earnings forecast revisions. Journal of Accounting Research 45 (3): 481513.
  • Bova, F., Y. Dou, and O.-K. Hope. 2013. Employee ownership and firm disclosure. Working paper, University of Toronto and New York University.
  • Bradshaw, M. T. 2011. Analysts' forecasts: What do we know after decades of work? Working paper, Boston College.
  • Brochet, F., P. Naranjo, and G. Yu. 2012. Causes and consequences of linguistic complexity in non-U.S. firm conference calls. Working paper, Harvard University and MIT.
  • Brown, L. D., and K. Huang. 2013. Forecast-recommendation consistency and earnings forecast quality. Accounting Horizons 27 (3): 45167.
  • Bryan, S. H. 1997. Incremental information content of required disclosures contained in management discussion and analysis. The Accounting Review 72 (2): 285301.
  • Caskey, J. A. 2009. Information in equity markets with ambiguity-averse investors. Review of Financial Studies 22 (9): 3595627.
  • Clement, M. B. 1999. Analyst forecast accuracy: Do ability, resources, and portfolio complexity matter? Journal of Accounting and Economics 27 (3): 285303.
  • Clement, M. B., and S. Tse. 2003. Do investors respond to analysts' forecast revisions as if forecast accuracy is all that matters? The Accounting Review 78 (1): 22749.
  • Cooper, R., T. Day, and C. Lewis. 2001. Following the leader: A study of individual analysts' earnings forecasts. Journal of Financial Economics 61 (3): 383416.
  • Cready, W. M., and D. N. Hurtt. 2002. Assessing investor response to information events using return and volume metrics. The Accounting Review 77 (4): 891909.
  • Dambra, M., C. E. Wasley, and J. S. Wu. 2013. Soft-talk management cash flow forecasts: Bias, quality, and stock price effects. Contemporary Accounting Research 30 (2): 60744.
  • D'Souza, J., K. Ramesh, and M. Shen. 2010. The interdependence between institutional ownership and information dissemination by data aggregators. The Accounting Review 85 (1): 15993.
  • De Franco, G., and O.-K. Hope. 2011. Do analysts' notes provide new information? Journal of Accounting, Auditing and Finance 26 (2): 22954.
  • De Franco, G., F. P. Vasvari, D. Vyas, and R. Wittenberg-Moerman. 2013. Debt analysts' views of debt-equity conflicts of interest. Working paper, University of Toronto, London Business School, and University of Chicago.
  • De Franco, G., and Y. Zhou. 2009. The performance of analysts with a CFA designation: The role of human-capital and signaling theories. The Accounting Review 84 (2): 383404.
  • Emery, R., and X. Li. 2009. Are the Wall Street analyst rankings popularity contests? Journal of Financial and Quantitative Analysis 44 (2): 41137.
  • Feldman, R., S. Govindaraj, J. Livnat, and B. Segal. 2010. Management's tone change, post earnings announcement drift and accruals. Review of Accounting Studies 15 (4): 91553.
  • Francis, J., D. J. Nanda, and P. Olsson. 2008. Voluntary disclosure, earnings quality, and cost of capital. Journal of Accounting Research 46 (1): 5399.
  • Gleason, C. A., and C. M. C. Lee. 2003. Analyst forecast revisions and market price discovery. The Accounting Review 78 (1): 19325.
  • Harris, M., and A. Raviv. 1993. Differences in opinion make a horse race. Review of Financial Studies 6 (3): 47394.
  • Healy, P. M. 1977. Can you understand the footnotes to financial statements? Accountants Journal 56: 21922.
  • Holthausen, R. W., and R. E. Verrecchia. 1988. The effect of sequential information releases on the variance of price changes in an intertemporal multi-asset market. Journal of Accounting Research 26 (1): 82106.
  • Holthausen, R. W., and R. E. Verrecchia. 1990. The effect of informedness and consensus on price and volume behavior. The Accounting Review 65 (1): 191208.
  • Hope, O.-K., W. B. Thomas, and G. Winterbotham. 2008. Geographic earnings disclosure and trading volume. Journal of Accounting and Public Policy 28 (3): 16788.
  • Huang, A., A. Zang, and R. Zheng. 2012. Large sample evidence on the informativeness of text in analyst reports. Working paper, Hong Kong University of Science and Technology.
  • Institutional Investor. 2010. The 2010 all-America research team. October: 52.
  • Jacob, J., T. Z. Lys, and M. Neale. 1999. Expertise in forecasting performance of security analysts. Journal of Accounting and Economics 28 (1): 5182.
  • Kadous, K., L. Koonce, and K. L. Towry. 2005. Quantification and persuasion in managerial judgment. Contemporary Accounting Research 22 (3): 64386.
  • Kandel, E., and N. D. Pearson. 1995. Differential interpretation of public signals and trade in speculative markets. Journal of Political Economy 103 (4): 83172.
  • Karpoff, J. M. 1987. The relation between price changes and trading volume: A survey. Journal of Financial and Quantitative Analysis 22 (1): 10926.
  • Kim, O., and R. E. Verrecchia. 1991. Trading volume and price reactions to public announcements. Journal of Accounting Research 29: 30221.
  • Kim, O., and R. E. Verrecchia. 1994. Market liquidity and volume around earnings announcements. Journal of Accounting and Economics 17 (1–2): 4167.
  • Kim, O., and R. E. Verrecchia. 1997. Pre-announcement and event-period private information. Journal of Accounting and Economics 24 (3): 395419.
  • Kothari, S. P., X. Li, and J. E. Short. 2009. The effect of discourses by management, analysts, and business press on cost of capital, return volatility, and analyst forecasts: A study using content analysis. The Accounting Review 84 (5): 163970.
  • Lawrence, A. 2012. Individual investors and financial disclosure. Working paper, University of California at Berkeley.
  • Li, F. 2008. Annual report readability, current earnings, and earnings persistence. Journal of Accounting and Economics 45 (2–3): 22147.
  • Li, H. 2013. Repetitive disclosures in the MD&A. Working paper, University of Toronto.
  • Mikhail, M., B. Walther, and R. Willis. 1997. Do security analysts improve their performance with experience? Journal of Accounting Research 35 (Supp.): 13157.
  • Mikhail, M. B., B. R. Walther, and R. H. Willis. 2003. The effect of experience on security analyst underreaction. Journal of Accounting and Economics 35 (1): 10116.
  • Mikhail, M. B., B. R. Walther, and R. H. Willis. 2007. When security analysts talk, who listens? The Accounting Review 82 (5): 122753.
  • Mozes, H. A. 2003. Accuracy, usefulness and the evaluation of analysts' forecasts. International Journal of Forecasting 19 (3): 41734.
  • Preston, A. M., and J. J. Young. 2000. Constructing the global corporation and corporate constructions of the global: A picture essay. Accounting, Organizations and Society 25 (4–5): 42749.
  • Previts, G. J., R. J. Bricker, T. R. Robinson, and S. J. Young. 1994. A content analysis of sell-side financial analyst company reports. Accounting Horizons 8: 5570.
  • Ramnath, S., S. Rock, and P. S. Shane. 2008. Financial analysts' forecasts and stock recommendations: A review of the research. Foundations and Trends in Finance 2 (4): 311420.
  • Rogers, R. K., and J. Grant. 1997. Content analysis of information cited in reports of sell-side financial analysts. Journal of Financial Statement Analysis 3 (1): 1730.
  • Rogers, J. L., A. Van Buskirk, and S. L. Zechman. 2011. Disclosure tone and shareholder litigation. The Accounting Review 86 (6): 215583.
  • SEC. 1998. A plain English handbook: How to create clear SEC disclosure documents. Washington, DC: SEC.
  • Smith, J. E., and N. P. Smith. 1971. Readability: A measure of the performance of the communication function of financial reporting. The Accounting Review 46 (3): 55261.
  • Stickel, S. E. 1990. Predicting individual analyst earnings forecasts. Journal of Accounting Research 28 (2): 40917.
  • Stickel, S. E. 1992. Reputation and performance among security analysts. Journal of Finance 47 (5): 181136.
  • Tetlock, P. C. 2007. Giving content to investor sentiment: The role of media in the stock market. Journal of Finance 62 (3): 113968.
  • Tetlock, P., M. Saar-Tsechansky, and S. Macskassy. 2008. More than words: Quantifying language to measure firms' fundamentals. Journal of Finance 63 (3): 143767.
  • Verrecchia, R. E. 2001. Essays on disclosure. Journal of Accounting & Economics 32 (1–3): 97180.