The effect of industry co-location on analysts’ information acquisition costs Jared Jennings Olin Business School Washington University in St. Louis St. Louis, MO 63130-6431 [email protected] Joshua Lee Olin Business School Washington University in St. Louis St. Louis, MO 63130-6431 [email protected] Dawn Matsumoto* Foster Business School University of Washington Seattle, WA 98133 [email protected] April 2014 *Corresponding author. We thank Kris Allee, Dave Burgstahler, Ted Christensen, Peter Demerjian, Weili Ge, Bret Johnson, Sarah McVay, Rick Mergenthaler, Mike Minnis, Terry Shevlin, and Jake Thornock for helpful comments. We also thank workshop participants at Santa Clara University, the University of Washington, the Nick Dopuch Conference at Washington University, and the Accounting Research Symposium at Brigham Young University. All errors are our own. Electronic copy available at: http://ssrn.com/abstract=2308879 The effect of industry co-location on analysts’ information acquisition costs ABSTRACT: We examine how the co-location of firms in the same industry affects analysts’ cost of gathering and processing information. Prior research finds evidence consistent with firms being more knowledgeable about other firms in the same geographical area. We argue that this improved knowledge base affects the information set of financial analysts. Consistent with this conjecture, we find that analysts include fewer firms in their portfolios when following firms located farther away from other firms in the same industry. In addition, we find that firms located farther away from other firms in the same industry have lower analyst following and less timely forecast revisions. We find that the additional costs that analysts incur to follow distant firms are amplified when earnings are more difficult to forecast. Lastly, we provide some evidence that managers communicate their knowledge about other firms in the same geographic area to analysts and investors. Specifically, during conference calls, managers are more likely to reference firms in their industry that are geographically closer. This paper provides additional evidence that the co-location of firms in the same industry not only affects operating and strategic decisions (as documented in the existing literature) but also analysts’ costs of gathering and analyzing information as well as managers’ communication with external market participants. Keywords: Geographic Location; Voluntary Disclosure; Security Analysts JEL Classification: D83; M40; M41; R10; R12 Electronic copy available at: http://ssrn.com/abstract=2308879 I. INTRODUCTION The co-location of firms in the same geographic area can affect the firm’s operations, strategy, and information environment. Marshall (1920) suggests that co-locating near other firms in the same industry results in several positive externalities, such as knowledge spillovers that facilitate innovation and growth, specialized labor forces, and reduced costs from obtaining inputs or shipping products to customers. The prior research suggests that these positive externalities increase managers’ knowledge and awareness of other firms in the same geographic area (e.g., Lovely et al., 2005; Audretsch and Feldman, 1996). We add to the literature by examining how the firm’s location can affect its information environment. Specifically, we examine whether financial analysts incur higher information gathering and processing costs when following firms located farther away from other firms in the industry. We predict that analysts’ information acquisition costs decrease when the firm is geographically closer to other firms in the industry for three reasons. First, managers’ increased knowledge of other firms in their geographical area, if communicated to analysts, is likely to improve analysts’ information set. The prior research provides evidence consistent with executives and employees being more knowledgeable about other firms in the same geographic area (Jaffe et al., 1993). Firms are also more likely to co-locate when the benefits of knowledge sharing are higher (Audretsch and Feldman, 1996; Lovely et al., 2005). We conjecture that managers communicate this information to analysts. For example, managers may compare or benchmark their firm’s operations and performance with other firms that are located in the same area. Second, analysts are more likely to visit firms that are located in close proximity to other firms in the industry, which likely provides informational benefits by allowing analysts 1) to speak informally with managers as well as with lower level employees and 2) to quickly and effectively compare the operations and facilities of the firms in the same geographic area. Third, local media outlets are more likely to compare and contrast local firms (i.e., firms in the same geographic area) reducing the analysts’ costs of performing similar analyses. 1 We conduct several tests to examine analysts’ costs of following firms that are located geographically farther away from other firms in the industry.1 We first find that analysts follow fewer firms when the average distance between the firms in the analyst’s portfolio and other firms in the same industry increases. While this result implies that analysts bear higher costs when following firms located farther from other firms in the industry, it does not necessarily imply that firms located farther from other firms in their industry attract lower analyst attention or that analysts obtain less information about these firms. Indeed, the benefits from following these distant firms might outweigh the greater information acquisition costs such that analysts reduce their effort or coverage of other firms in the industry. We specifically test whether the firm’s location in relation to other firms in the industry affects analyst following and the quality of the analysts’ output. We find evidence that analyst following is lower for firms that are located farther away from other firms in the same industry. We also find that analysts revise their forecasts in a less timely manner when following geographically more distant firms, suggesting that analysts incur higher information acquisition and processing costs when following these firms. Interestingly, we do not find evidence in our main results that analyst forecast accuracy is lower, on average, for firms located farther away from other firms in the industry. We also examine when the geographical spacing of the firm relative to other firms in the industry has a greater influence on analyst following, the timeliness of analyst revisions, and analyst forecast accuracy. We expect the additional costs that analysts incur to follow firms farther from other firms in the same industry are less relevant when earnings are relatively easy to forecast, potentially offsetting the negative effects of being located farther away. First, we find that the negative association between the size of the analyst’s portfolio and the average distance between the firms included in the analysts’ portfolio and other firms in the same industry is magnified when earnings are more difficult to forecast. We also 1 While a firm’s choice of location is not exogenously determined, we do not believe that effects on analysts’ costs play a firstorder role in management’s decision on where to locate. Thus, we do not believe that reverse causality is a plausible alternative explanation. It is possible that a third factor affects both the decision to locate outside an industry cluster as well as analyst costs (i.e., a possible correlated omitted variable problem). As we discuss in our research design, we include numerous control variables that prior research has identified as determinants of our dependent variables, as well as variables that capture potential factors that determine the location decision. In addition, as discussed further below, our hypotheses include cross-sectional analyses that reduce the likelihood that our results are driven by a correlated omitted variable. 2 find that analyst following, revision timeliness, and even forecast accuracy are more negatively associated with the firm’s relative distance to other firms in the industry when earnings are inherently more difficult to forecast. These results suggest that forecasting difficulty exacerbates the costs of acquiring and analyzing information for firms that are geographically distant from other firms in the same industry. Importantly, these results hold while controlling for several known determinants of the costs and benefits of information acquisition, including size, growth, institutional ownership, performance, share turnover, volatility, and business concentration. We also control for the co-movement of earnings, as Engelberg et al. (2010) argue that the geographical clustering of firms within an industry leads to greater co-movement of fundamentals. Our results suggest that the co-movement of fundamentals associated with the firm’s geographic proximity to other firms in the industry is not the only mechanism through which distance affects analysts’ information acquisition costs. Rather, managers’ knowledge of other firms in the industry also likely plays a role. Lastly, we provide some evidence to support our conjecture that managers communicate a portion of their information about other firms in the same geographic area to analysts and other market participants. Specifically, we examine the extent to which managers reference other firms during their conference calls. We provide evidence that the probability of a manager mentioning a firm in the same industry during a conference call decreases as the geographic distance between the two firms increases. This evidence is consistent with managers having a greater knowledge of other firms in the same industry and geographical area and with managers communicating this information to investors and analysts. We contribute to both the analyst literature as well as the literature on firm geography. Beyer et al. (2010) suggests that more research is needed to understand why firms are added to analyst portfolios and what firm characteristics affect this decision. We directly add to the analyst literature by providing evidence that the firm’s location can affect analyst information acquisition costs, thereby affecting analyst portfolio composition decisions, the number of analysts following a firm, analysts’ ability to quickly provide information to investors, and the accuracy of analyst outputs. We also provide evidence that the additional analyst costs associated with locating farther away from other firms in the industry are 3 primarily driven by firms with earnings that are more difficult to forecast. This evidence is important to investors and managers in understanding the costs that analysts incur to follow firms. We also contribute to the growing literature on firms’ geographical location by examining the effects of a firm’s location relative to other firms in its industry. Prior studies have examined an individual analyst’s geographical location relative to the firm and the impact this has on that individual analysts’ forecasting behavior (e.g., Orpurt, 2004; Malloy, 2005; Bae et al., 2008). In addition, Loughran and Schultz (2005) find that fewer analysts follow firms located in rural areas, which are unlikely to be areas of high analyst concentration. We believe that we make an important contribution to the literature by documenting how the firm’s relative location to other firms in the industry affects the firm’s information environment. Also, in additional robustness tests, we control for whether the firm is located in a rural or urban area and find quantitatively similar results. Thus, our results do not seem to be explained by the firm’s location in an area of high analyst concentration. Engelberg et al. (2010) also examine the effects of a firm’s location relative to other firms in their industry. They demonstrate greater co-movement in fundamentals among firms in an industry cluster, leading to higher analyst following. We demonstrate that the effect of geographical clustering on analyst following go beyond the co-movement of fundamentals, as the effect of distance relative to others in the industry exists even after controlling for earnings co-movement. Moreover, our evidence based on our analysis of management’s communication with analysts during conference calls supports our theory that managers have more information about geographically closer firms in the industry and that they communicate this information to analysts, improving the information set of analysts. Understanding how the firms’ location can influence the firm’s information environment is likely of importance to managers, investors and other market participants. The rest of the paper is organized as follows. In Section 2, we discuss the prior literature and develop our hypotheses. In Section 3, we outline the empirical models, describe the sample and discuss the results of the tests of our hypotheses. In Section 4, we examine the effect of co-location on managers’ voluntary disclosure policies. In Section 5, we conduct several robustness tests. We conclude this study in Section 6. 4 II. THEORY AND HYPOTHESIS DEVELOPMENT Co-location Marshall (1920) identifies three positive externalities that arise from the localization of an industry. First, the localization of an industry creates a demand for specialized labor. Second, the geographic clustering of firms in an industry likely reduces the costs of obtaining inputs from suppliers and shipping goods to customers. Third, the geographic clustering of firms within an industry allows for knowledge spillovers that facilitate innovation and growth among firms in the industry. These positive externalities likely improve managers’ knowledge of other firms in the same industry and geographical area. For example, the demand for specialized labor can result in a shared labor pool and employees moving between companies in the same industry. Prior research supports the notion that managers are more knowledgeable of firms in close geographic proximity. For example, Jaffe et al. (1993) find evidence that patent citations are attenuated when moving farther away from the original patent location. Prior research also provides evidence that firms co-locate when the benefits to knowledge sharing are greater. For example, Lovely et al. (2005) find evidence that firms dealing in high levels of foreign market exports tend to co-locate in the same geographic location, especially when foreign market information is difficult to obtain. Audretsch and Feldman (1996) find that more research and development intensive firms are more likely to co-locate in similar geographic areas. The idea that firms in the same industry, particularly competitors, would willingly share information with each other might seem implausible. However, Stein (2008) provides a theoretical framework for information spillovers among competitors. He argues that firms mutually benefit from conversations with their competitors and analytically shows that honest knowledge sharing can occur between firms even if the firms are in competition, as long as there are complementarities in idea 5 production – that is, that firms build on and refine an idea when exchanging information.2 Moreover, knowledge sharing can occur not just with competitors but also with other firms in a firm’s supply chain – i.e., customers and suppliers. Analyst Behavior We consider how the co-location of firms within an industry affects the information acquisition and processing costs of an important set of information intermediaries, financial analysts. Investors use analyst research (e.g., earnings forecasts, stock recommendations, target prices, and other qualitative and quantitative analyses) to help establish earnings expectations and aid in the price discovery process (e.g., Brown and Rozeff, 1978; Givoly and Lakonishok, 1979; Brown et al, 1987; Fried and Givoly, 1982; Asquith et al., 2005; Frankel et al., 2006). Therefore, understanding the factors that impact the quality of analyst output is important. Analysts gather data from many sources including firm disclosures, industry publications, market trends, and informal communication with the firm’s employees. We anticipate that the geographic location of the firm relative to other firms in the industry reduces analysts’ costs of obtaining and analyzing useful and relevant information in three ways. First, as discussed above, executives and employees likely know more about other firms in the same geographic area. To the extent managers communicate this information to analysts (as hypothesized above) and this information is useful to analysts, following firms located in close proximity to other firms in the industry will improve the analysts’ information set. For example, managers or employees may use other firms as a benchmark or reference point when talking about their firm. Managers can communicate this information to analysts publicly (e.g., during a firm-sponsored conference call) as well as privately (e.g., during a site visit or a 2 Despite the benefits, not all firms choose to locate in close proximity to their competitors, suggesting additional costs or reduced benefits to being located in an industry cluster. Shaver and Flyer (2000) argue that the costs of contributing to the industry cluster outweigh the benefits of being part of the cluster for firms with the best technologies, training programs, suppliers, or distributors. As a result, these firms are more likely to locate outside the industry cluster because they are less likely to benefit from knowledge spillovers, a specialized labor force, and reduced transportation costs. Alcacer (2006) provides evidence consistent with more-capable firms co-locating less often than less-capable firms in the cellular handset industry. Close geographic locations can also increase competition in some industries. Baum and Haveman (1997) argue that hoteliers in Manhattan locate close to each other when they are similar on one product dimension but different on other dimensions to avoid local competition. 6 one-on-one phone call). While management cannot provide any private and material information that is not also disclosed to the public (due to Regulation Fair Disclosure), analysts can ask clarifying and follow-up questions about the firm’s current or future operations to improve the information content of their recommendations or the accuracy of their forecasts.3 Second, when firms in the same industry are located in close proximity to one another, analysts may be more likely to visit these firms and more likely to visit multiple firms in quick succession. While out-of pocket transportation costs are admittedly trivial for most brokerage houses, time constraints are likely real and, at the margin, could affect an analysts’ propensity to make a site visit. Physically visiting a firm’s location likely provides additional informational benefits because it allows the analyst to talk with managers informally and also allows them to speak with lower level employees. 4 In addition, visiting multiple locations in quick succession likely enables analysts to more efficiently compare and contrast the operations of those firms and their conversations with management and other employees, potentially allowing them to identify subtle differences in the firms’ operations. Finally, local information sources (e.g., local newspapers) may reduce analysts’ costs of gathering and analyzing data about related firms located in the same geographical location. Local information sources are more likely to compare and contrast the operations and performance of related firms in their geographically area, reducing the amount of data gathering and analysis performed by the analyst. Based on the above, we predict that the costs of gathering and processing information about a firm are significantly higher for firms that are geographically more distant from other firms in the same industry. We formally state our first hypothesis in alternative form below. H1 - Analysts incur greater costs to follow firms that are geographically farther away from other firms in the same industry. 3 As an example, in a recent conference call held by Herbalife (January 10, 2013), management openly invited analysts and other interested market participants to call management and visit the production facilities to obtain clarification on the firm’s business model. In addition, Green et al. (2012) provide evidence that analysts who interact with corporate managers at broker-hosted conferences have more informative stock recommendations and this persists in the post Reg FD era. 4 As evidence of the information advantage of visiting a firm’s physical location, Malloy (2005) finds that individual analysts who are located closer to firms (presumably allowing them to visit the physical location more frequently) are relatively more accurate than other analysts following the firm. 7 To the extent managers are unwilling to share information about other firms in the same geographical area, site visits are ineffective, and the local information sources are uninformative, the geographic distance of the firm to other firms in the same industry may not affect the information acquisition costs of following a firm. It is possible that the vast majority of analysts’ information is gathered through other channels such as mandatory disclosure, industry publications, and other industry and firm specific data sources. Also, if firms operating farther away from other firms in the same industry have earnings that are easier for analysts to forecasts, this could offset the increased costs and reduced benefits from being located farther away. As a result, we examine how forecasting difficulty affects the relation between the firm’s geographical spacing and analyst costs. We expect earnings that are easier to forecast to at least partially offset the higher costs associated with following firms that are geographically farther away from other firms in the same industry. Our formal hypothesis stated in alternative form is: H2 - Analysts’ information acquisition costs of following a firm that is geographically farther away from other firms in the same industry increase with forecasting difficulty. III. EMPIRICAL ANALYSIS Sample Selection We obtain our sample from the universe of U.S. firms with data from Compustat, CRSP, and IBES between 1994 and 2011.5 We exclude firms in the financial services (SIC 6000 to 6999) and utilities (SIC 4900 to 4999) industries due to differences in the information environment of these regulated industries. We also require at least 20 observations per year for each two-digit SIC code to ensure that firms have the opportunity to locate near other firms in the same industry.6,7 We also require there to be sufficient data to calculate the dependent and independent variables in our regression analyses. 5 We choose 1994 as the initial start date of our sample because we collect address information from company reports issued to the SEC which are readily available online through EDGAR beginning in 1994. 6 Our results are robust to alternative industry cutoffs of 10, 30, and 50 firms. 8 Following prior research, we use the firm’s headquarters as a proxy for its location.8 We collect the addresses of firms’ headquarters from reports filed with the SEC for each firm/year.9 When missing, we use the addresses provided by Compustat. We measure the distance of a firm relative to all other firms in its industry using the universe of firms for which we have location data, regardless of the availability of data to estimate our regression analyses. We calculate two variables to estimate the distance of a firm to other firms within its industry.10 First, we define DIST 10i,t as the mean distance between the firm’s headquarters and the headquarters of the 10 closest firms within the industry (two-digit SIC code) for each year. Second, we define # FIRMS 100i,t as the number of firms in the industry located within 100 miles of the firm in each year.11,12 We multiply # FIRMS 100i,t by negative one so that DIST 10i,t and # FIRMS 100i,t are both increasing in the geographic spacing of firms within the industry. We then 7 We define industry based on 2-digit SIC industry code classification to allow for the possibility of suppliers and customers being included in the same industry, which we consider an important source of the information spillover effect (i.e., knowledge about significant suppliers and customers that are communicated to analysts likely helps the analyst in forecasting for these firms). An alternative measure of industry based on Hoberg and Phillips (2010) primarily identifies competitors, which is why we do not choose to use their definitions for our main tests. However, we test the sensitivity of our results to defining our distance variables using their industry definitions in Section 5. 8 Several finance and management papers use the firm’s headquarters as its location such as Coval and Moskowitz (1999), Zhu (2002), Ivkovic and Weisbenner (2005), Loughran and Schultz (2004 and 2005), Malloy (2005) and Chhaochharia, Kumar, and Niessen-Ruenzi (2012). We use the headquarters location rather than the state of incorporation because firms tend to incorporate in states with favorable tax and bankruptcy laws and often do not incorporate in the state of their primary operations (Coval and Moskowitz 1999). The headquarters is likely close to the firm’s principal operations and the headquarters is also where management operates. For these reasons, information transfers are most likely to occur in the geographic area of the firm’s headquarters. However, to the extent a firm operates in multiple geographic locations, information spillovers may occur in areas outside the firm’s headquarters. 9 Compustat provides the most recent address of the firm. However, because our sample period covers such a long time span, we manually collect the address of the firm’s headquarters as of each reporting date. The address is found in the firms’ annual report and is potentially different than the business and mailing addresses provided in the header of each filing. We identify the headquarters as the address the company specifically specifies as its “principal executive offices.” 10 Chhaochharia et al. (2012) use similar methods to measure the distance of the firm to institutional investors. 11 We choose to define our distance measures using the number rather than the percentage of firms in the industry located near the firm’s headquarters. Information spillovers are likely to occur for firms located nearby, regardless of the number of other firms in the industry that are not located nearby. For example, consider two firms, firm A and firm B. Firm A has 5 firms located nearby and 20 firms in its industry. Firm B has 25 firms located nearby and 100 firms in its industry. An industry concentration measure would indicate that 25 percent of firms are located nearby for both firms. However, firm B has greater potential for information spillovers given the greater number of firms located nearby. For this reason, we use the number rather than the percentage of firms located nearby in our main analysis. As a robustness test, we also examine distance measures using the percentage of firms within an industry that are located nearby. See section 5 for additional detail. 12 As an additional robustness test, we perform all analyses using the mean distance between the firm and the 5 closest firms in its industry and the number of firms located within 50 miles and find qualitatively similar results for all tests (untabulated). See section 5. 9 standardize the distance measures to take values between 0 and 1 by placing them in deciles from 0 to 9 and dividing by 9.13 We next examine how several firm specific characteristics vary with our measures of distance. In Table 1 we report the mean values of various firm characteristics for the full sample and for the partitions of firms in the top and bottom halves of the distance distribution (computed using the DIST 10i,t variable). In the final column, we report the p-values testing the difference in mean values between the firms in the top and bottom partitions. We note no difference in market capitalization for firms that are farther away from other firms in the same industry (MKT VALi,t). We find that firms that are farther away from other firms in the industry have higher book-to-market ratios (BTMi,t), are more profitable (ROAi,t), have lower stock turnover (TURNi,t), are less concentrated in a single industry (BUS CONCi,t), have lower earnings volatility (EARN VOLi,t), lower return volatility (RET VOLi,t), are older (AGEi,t), and have earnings that are more likely to co-move with the earnings of the ten geographically closest firms in the industry (EARN COMOVEi,t) (see Appendix A for variable definitions).14, These significant differences highlight the need to control for firm specific characteristics when examining our hypotheses. In untabulated results we examine the differences between the bottom and top distance partition using the # FIRMS 100i,t variable and find qualitatively similar results. Table 1 also reports the average values of our distances measures for the full sample and for high and low levels of DIST 10i,t. The values indicate that on average the firms in the top half of DIST 10i,t are located approximately 229 miles from the ten closest firms in the same industry and have 4 industry firms 13 As examples, the following firms fell in the upper deciles of our distance measures. First, Key Tronic Corp, an electronics manufacturing company located in Spokane, Washington is approximately 750 miles from the majority of the firms in its industry (industrial and commercial machinery and computer equipment), which are located in San Jose, CA. Another example is Tesla Motors Inc. located in Palo Alto, California which is over 2,000 miles from several major competitors in Detroit, Michigan (e.g., Ford, GM, Spartan Motors). Finally, U.S. Energy Corp is located in Wyoming whereas the majority of the firms in its industry (oil and gas extraction) are located in Texas. 14 The finding that firms located farther from firms in their industry have earnings that co-move more with the closest firms in their industry is somewhat contrary to the findings in Engelberg et al. (2010). However, if we use industry-adjusted variables, we find negative association between our distance measures and earnings co-movement, consistent with the findings in Engelberg et al. (2010). Our main analysis includes industry fixed effects, which is equivalent to using industry mean-adjusted variables. 10 within 100 miles. In contrast, firms in the bottom half of the distribution are located approximately 10 miles from the ten closest firms in the same industry and have 56 firms within 100 miles. Tests of Hypothesis 1 We use several tests to examine whether the analysts’ costs of gathering and analyzing information about the firm increase when the firm is farther away from other firms in the industry. We start by examining whether the number of firms included in the analyst’s portfolio decreases when the firms in the analyst’s portfolio are farther away from other industry firms. We then move to firm specific analyses and test whether analyst following, forecast revision timeliness, and accuracy are affected by the location of the firm relative to other firms in the industry. Analyst Portfolios To examine whether the number of firms included in analysts’ portfolios is smaller when the average distance of the firms in the portfolio is higher, we run the below regression by analyst/year: ln(# FIRMS FOLLa,t) = α0 + α1 MEAN DISTa,t + α2 BROKER SIZEa,t + (1) α3 ln(EXPERIENCEa,t) + α4 ln(# INDUSTRIESa,t) + α5 MEAN ln(MKT VAL)a,t + α6 MEAN BTMa,t + α7 MEAN INST OWNa,t + α8 MEAN ROAa,t + α9 MEAN TURNa,t + α10 MEAN BUS CONCa,t + α11 MEAN EARN VOLa,t + α12 MEAN RET VOLa,t + α13 MEAN ln(AGE)a,t + α14 MEAN EARN COMOVEa,t + εa,t The # FIRM FOLLa,t, variable is equal to the number of firms analyst a follows during year t. The MEAN DISTa,t variable is the primary coefficient of interest and is either equal to the MEAN DIST 10a,t or MEAN # FIRMS 100a,t variable. The MEAN DIST 10a,t (MEAN DIST 10a,t) variable is equal to the average DIST 10i,t (# FIRMS 100i,t) variable for the firms that analyst a follows during year t. We then rank the average distance variable into deciles from 0 to 9, and divide by 9. We include several control variables that potentially explain the size of an analysts’ portfolio to reduce the likelihood that correlated omitted variables are biasing our inferences. BROKER SIZEa,t is a proxy for the size of the brokerage house for which analyst a works, measured using the number of firms 11 followed by the brokerage house during year t. We standardize the BROKER SIZEa,t variable to take on values between 0 and 1 by taking an ordinal ranking of the brokerage houses each year and dividing by the total number of brokerage houses. We expect larger brokerage houses to have greater resource support (e.g., subordinate analysts, databases, estimation techniques industry connections, etc.), which likely allows for faster gathering and processing of information, resulting in larger portfolios. We also control for the analyst’s experience (EXPERIENCEa,t), which is equal to the number of years that analyst a is found on IBES as of the end of year t. More experienced analysts are potentially able to handle larger portfolios. Finally, we control for the number of industries (# INDUSTRIESa,t) followed by the analyst, defined as the number of unique industries (two-digit SIC) followed by analyst a during year t. To the extent covering multiple industries is more difficult, we expect analysts to have smaller portfolios when covering more industries. We also control for certain characteristics of the firms in the analyst’s portfolio that are potentially correlated with our distance measures. Using the variables identified in Table 1, we compute the mean of each variable for all firms in analyst a’s portfolio at time t and re-label the variables with the term MEAN in front to differentiate from the firm-specific variables reported in Table 1. For example, for return on assets we compute the mean ROAi,t for all firms in analyst a’s portfolio at time t and call the resulting variable MEAN ROAa,t. We include controls for the means of ln(MKT VAL)i,t, BTMi,t, INST OWNi,t, ROAi,t, TURNi,t, BUS CONCi,t, EARN VOLi,t, RET VOLi,t, ln(AGE)i,t, and EARN COMOVEi,t. In addition to the above, we include year indicator variables to control for unidentified time-variant fluctuations in the size of analysts’ portfolios. We also cluster the standard errors by analyst to correct the standard errors for potential serial-correlation. Table 2 presents the results using the MEAN DIST 10i,t and MEAN # FIRMS 100a,t variables in Column 1 and 2, respectively. We find that the coefficients on the MEAN DIST 10a,t (MEAN # FIRMS 100i,t) variable is equal to -0.191 (-0.455) and significant at the 1% level. Since we take the natural log of the dependent variable, we are able to interpret the change from the 1st to 10th decile (values of 0 to 1) in the MEAN DISTa,t variable as a percentage change in the dependent variable. As a result, the coefficient 12 of -0.191 (-0.455) suggests a 19% (46%) decrease in the number of firms followed by the analyst when the MEAN DIST 10a,t (MEAN # FIRMS 100a,t) variable moves from the 1st to 10th deciles, suggesting an economically meaningful effect of distance on the size of an analyst’s portfolio. We find that the coefficient on the BROKER SIZEa,t variable is positive and significant, suggesting that analysts employed by larger brokerage houses have more resources and are able to follow more firms. The coefficient on the EXPERIENCEa,t variable is positive and significant, suggesting that experienced analysts follow more firms. The positive coefficient on the # INDUSTRIESa,t variable is contrary to our expectations; however, it is possible that “generalist” analysts (i.e., analysts who do not specialize in a particular industry) cover more firms than “specialist” analysts. The coefficients on our average firm variables suggest that analysts cover more firms when the firms in their portfolio have greater business segment concentration, lower return volatility, and greater earnings co-movement with other firms in the geographic area and industry consistent with lower costs of covering these firms. We also find a negative coefficient on the mean institutional ownership variable suggesting that analysts follow fewer firms when the firms in their portfolios have a greater information demand from large shareholders. These results are consistent with the notion that covering more distant firms is more costly to analysts, resulting in smaller portfolio sizes. Whether this additional cost results in distant firms having lower coverage depends on whether the benefits of covering these firms are higher, which we explore in the next sub-section.15 Analyst Following Our first firm-level test examines whether analyst following is lower for geographically more distant firms. The Pearson and Spearman correlations between the natural log of ANAL FOLLi,t and the 15 While the analysts’ portfolio test sheds light on whether analysts bear greater costs when covering distant firms, it does not necessarily imply that more distant firms are covered by fewer analysts. Firm-specific benefits may offset the costs of following more distant firms and analysts may incur the costs to follow these firms. This would imply lower overall analyst coverage in the economy but does not necessarily imply that more distant firms will be followed less often. We, therefore, specifically test whether firms that are farther away from other firms within the industry attract less analyst following. 13 distance measures (DIST 10i,t and # FIRMS 100i,t) are indeed negative and statistically significant, ranging between -0.05 and -0.07 (untabulated). These correlations, while small, provide preliminary evidence that analyst following is lower when the firm is located farther away from other firms in the industry. We now turn our attention to the multivariate regression results. Our multivariate regression is defined in Equation 2. ln(ANAL FOLL)i,t = α0 + α1 DISTANCEi,t + α2 ln(MKT VALi,t) + α3 BTMi,t + α4 INST OWNi,t + α5 (2) ROAi,t + α6 TURNi,t + α7 BUS CONCi,t + α8 EARN VOLi,t + α9 RET VOLi,t + α10 ln(AGEi,t) + α11 EARN COMOVEi,t + YEAR + INDUSTRY + εi,t All dependent and independent variables are as previously defined or defined in Appendix A. In addition, we include year and industry (two-digit SIC code) indicator variables to control for systematic shifts in analyst following over time and across industries. We cluster the standard errors by firm due to potential serial correlation in our dependent and independent variables (Petersen 2009). For all subsequent tests, we include industry and year indicators variables and cluster the standard errors by firm unless specified differently. The DISTANCEi,t variable, defined as either DIST 10i,t or # FIRMS 100i,t, is our main variable of interest. Column 1 (2) of Table 3 includes the results when the DISTANCEi,t variable is equal to the DIST 10i,t (# FIRMS 100i,t) variable. We find that the coefficients on the DIST 10i,t and # FIRMS 100i,t variables are significant at the 1% level and 10% level, respectively, and equal to -0.075 and -0.042, which suggests that moving from the 1st to 10th distance decile reduces analyst following by 4% to 8%. Similar to Bhushan (1989), we find a positive coefficient on the ln(MKT VALi,t) variable, suggesting that larger firms have higher analyst following. We find a positive coefficient on the book-tomarket ratio (BTMi,t), suggesting that lower growth or more established firms are more likely to have higher analyst following. Similar to Bhushan (1989) and Alford and Berger (1999), we find a positive coefficient on both the INST OWNi,t and TURNi,t variables, consistent with analysts’ responding to 14 investors’ demand for information. Contrary to Hayes (1998), we find a negative coefficient on ROAi,t.16 We find a positive coefficient on the business segment concentration variable (BUS CONCi,t), consistent with analysts being more likely to follow firms that concentrate in a specific industry (Gilson et al., 2001). We find a negative coefficient on the EARN VOLi,t and RET VOLi,t variables, suggesting that analysts are attracted to firms with less volatile operating environments. We find a negative coefficient on AGEi,t suggesting younger firms attract greater analyst following.17 Lastly, we find a positive coefficient on EARN COMOVEi,t suggesting that firms that respond similarly to local information shocks as other industry firms in their geographic area attract greater analyst following. This result is consistent with findings in Engelberg et al. (2010). However, the fact that our distance measures are negative and significant even after controlling for earnings co-movement suggests that the effect of distance on analyst following is not solely because the co-movement of fundamentals lowers analysts’ information acquisition costs. Forecast Revision Timeliness We next examine the timeliness of analyst revisions to further test whether analysts’ costs are higher when following a firm farther away from other firms in the industry. Similar to Zhang (2008), we assume that higher information acquisition and processing costs result in longer delays in issuing forecasts following important news events such as earnings announcements. QUICK REVi,t is equal to the percentage of analysts following firm i that issue a forecast revision within two days following the quarterly earnings announcement.18 To be consistent with our annual tests, we compute the average of the QUICK REVi,t variable over the four quarters during year t to obtain an annual measure of the timeliness of forecast revisions. The Pearson and Spearman correlations between the QUICK REVi,t and distance 16 The univariate correlation between analyst following and ROA is positive (0.16 and 0.17 for Pearson and Spearman correlations, respectively). Thus, the negative relation represents the marginal effect of performance on analyst following, after controlling for our other determinants of analyst following. 17 This result runs counter to our expectation. However, the univariate correlation between firm age and analyst following is positive (0.17). The reversed sign of the coefficient in the regression is likely due to the high correlation of firm age with other firm characteristics (e.g., firm size, institutional ownership, and volatility). 18 In addition to using the percentage variable, we also use an indicator variable equal to 1 if any analyst following firm i during year t issues a forecast revision within two days following the annual earnings announcement and find qualitatively similar univariate and multivariate results (untabulated) as those reported in Table 4. 15 variables (DIST 10i,t and # FIRMS 100i,t) are significantly negative at the 1% level, ranging between 0.066 and -0.068. This provides preliminary evidence that analysts are less likely to revise forecasts immediately following the annual earnings announcement when following firms farther away from other firms in the industry. In our multivariate analysis, we control for several firm characteristics, which are included in Equation 2. QUICK REVi,t = β0 + β1 DISTANCEi,t + β2 ln(MKT VALi,t) + β3 BTMi,t + β4 INST OWNi,t + β5 (3) ROAi,t + β6 TURNi,t + β7 BUS CONCi,t + β8 EARN VOLi,t + β9 RET VOLi,t + β10 ln(AGEi,t) + β11 EARN COMOVEi,t + β12 ln(ANAL FOLLi,t) + β13 BROKER SIZEi,t + β14 ln(# FIRMS FOLLi,t) + β15 ln(EXPERIENCEi,t) + YEAR + INDUSTRY + εi,t In addition to the control variables previously discussed, we also include three analyst specific variables: # FIRMS FOLLi,t, BROKER SIZEi,t, and EXPERIENCEi,t.19 # FIRMS FOLLi,t is the average number of firms covered by the analysts following firm i during year t. BROKER SIZEi,t is the average brokerage size for the analysts following firm i during year t. EXPERIENCEi,t is the average number of years of experience held by the analysts following firm i during year t. We expect a negative coefficient on # FIRMS FOLLi,t and positive coefficients on BROKER SIZEi,t and EXPERIENCEi,t.20 Table 4 reports the results using Equation 3. Similar to the preceding table, Column 1 (2) includes the results using the DIST 10i,t (# FIRMS 100i,t) as the main variable of interest. We find that the coefficients on the DIST 10i,t and # FIRMS 100i,t variables are equal to -0.022 and -0.023 and are 19 In previous tests, we show that analysts follow fewer firms when the firms in their portfolio are more distant from other firms in the industry suggesting greater costs of covering distant firms. To the extent analysts cover fewer firms when covering more distant firms, they would have more time to devote to the firms they do cover, potentially resulting in no difference in the timeliness to revision. We specifically control for this possibility by including these analyst specific variables. 20 In additional robustness tests, we include control variables for the percentage of firms in the industry that announce earnings in the [0,1] window surrounding the firm’s earnings announcement date and for the percentage of firms in the industry that have announced earnings as of the day prior to the firm’s earnings announcement date. If firms that co-locate near other firms in the industry are also more likely to time their announcements with other firms in the industry, our distance measures can potentially be picking up an industry information effect rather than a distance effect. We find evidence that our distance measures are negatively correlated with the percentage of concurrent announcements (correlation equal to -0.06) and positively correlated with the percentage of prior announcements (correlation equal to 0.03). Including these variables in Equation 3 does not affect the sign or significance of the coefficients on our distance variables. 16 significant at the 5% level. The coefficient estimate suggests moving from the lowest to the highest decile of the DIST 10i,t (# FIRMS 100i,t) variable reduces the percentage of analysts that quickly revise their forecasts by 5.1% (5.4%), relative to the mean of QUICK REVi,t. These results provide additional evidence that analysts incur higher costs when following firms located farther away from other firms in the industry. With respect to our control variables, we find analysts revise more quickly when institutional ownership (INST OWNi,t) and analyst following (ln(ANAL FOLLi,t)) are higher and performance (ROAi,t) and firm age (ln(AGEi,t)) are lower, suggesting that an increase in the demand for timely information increases with these factors. Analysts are also able to revise more quickly when firms have concentrated operations (BUS CONCi,t), suggesting that the costs of covering such firms are lower. This result corroborates Thomas (2002) who finds some evidence that analysts incur higher costs when following firms that are less operationally concentrated by industry. We also include controls for growth (BTMi,t), turnover (TURNi,t), earnings volatility (EARN VOLi,t), return volatility (RET VOLi,t), and earnings comovement (EARN COMOVEi,t); however, we find insignificant coefficients on each of these variables, with the exception of a negative coefficient on return volatility, suggesting that analyst forecast revisions are less timely when the firm’s operating environment is more volatile. With respect to our analyst variables, we find a positive and significant coefficient on BROKER SIZEi,t, suggesting that larger brokerage houses have more resources, which allow analysts to revise their forecasts more quickly. We also find that the coefficient on EXPERIENCEi,t is positive and significant, suggesting that more experienced analysts can more quickly process information and issue revisions. We do not find a significant coefficient on # FIRMS FOLLi,t. Analyst Accuracy Our last test of Hypothesis 1 examines the effect of the firm’s location relative to other firms in the industry on analysts’ forecast accuracy. To the extent analysts are unable to compensate for the lack of information obtained from firms that are located geographically farther away from other industry firms, analyst accuracy should suffer. We define ACCURACYi,t as the negative value of the absolute difference 17 between the last median quarterly analyst consensus forecast and actual EPS divided by share price for firm i in year t. Similar to the QUICK REVi,t variable, we compute the average of the ACCURACYi,t variable over the four quarters during year t to obtain an annual measure of accuracy. We find small but statistically positive Pearson and Spearman correlations between ACCURACYi,t and our distance measures (DIST 10i,t and # FIRMS 100i,t) of approximately 0.01, contrary to our expectations. However, we also find that our distance and accuracy measures are highly negatively correlated with both earnings volatility and return volatility. Thus, because more distant firms have lower volatility and lower volatility firms have higher forecast accuracy, the positive univariate correlations between our distance measures and accuracy might reflect a volatility effect rather than a distance effect. To examine whether this is the case, we estimate the following multivariate regression: ACCURACYi,t = β0 + β1 DISTANCEi,t + β2 ln(MKT VALi,t) + β3 BTMi,t + β4 INST OWNi,t + β5 (4) ROAi,t + β6 TURNi,t + β7 BUS CONCi,t + β8 EARN VOLi,t + β9 RET VOLi,t + β10 ln(AGEi,t) + β11 EARN COMOVEi,t + β12 ln(ANAL FOLLi,t) + β13 BROKER SIZEi,t + β14 ln(# FIRMS FOLLi,t) + β15 ln(EXPERIENCEi,t) + YEAR + INDUSTRY + εi,t Our results using Equation 4 are found in Table 5. Both coefficients on the DIST10i,t and # FIRMS 100i,t variables are insignificant, providing no evidence that analyst forecast accuracy is affected by the location of the firm relative to other firms in the industry after controlling for important characteristics associated with our distance measures. However, in the next sub-section we utilize additional cross-sectional tests to examine whether the relation between analyst accuracy and the firms location relative to other industry firms exists under certain conditions. We include the same control variables from Equation 3 to control for possible alternative explanations. Consistent with prior studies (e.g., Lang and Lundholm, 1996; Alford and Berger, 1999), we find positive and significant coefficients on ln(MKT VALi,t), TURNi,t, INST OWNi,t, and ROAi,t. We also find negative coefficients on EARN VOLi,t and RET VOLi,t, consistent with operating volatility making forecasting more difficult and/or reducing analysts’ incentives to gather information about the firm (Lang and Lundholm 1996). We find a negative coefficient on the BTMi,t, variable suggesting that 18 analysts are less accurate when following low growth firms. We also find a negative coefficient on the ln(AGEi,t) variable suggesting forecasts are more accurate for younger firms.21 With respect to the analyst variables, we find a positive and significant coefficient on the ANAL FOLLi,t variable (Alford and Berger, 1999), suggesting that higher analyst following results in a more accurate consensus forecast due to increased analyst competition. The coefficient on the ln(EXPERIENCEi,t) variable is negative and significant at the 10% level, which is contrary to our expectation. The coefficients on the BROKER SIZEi,t and # FIRMS FOLLi,t variables are insignificant. Test of Hypothesis 2 We now turn to our second hypothesis. We use the following regression to test whether the geographic spacing of the firm relative to other firms in the industry has a more negative effect on analysts’ information acquisition costs (Hypothesis 2). ANAL BEHAVIORt = η0 + η1 DISTANCEt + η2 HIGH EARN VOLt + η3 DISTANCEt * HIGH (5) EARN VOLt + CONTROLSt + εt We examine the four aspects of analyst behavior (ANAL BEHAVIORt) used to examine the first hypothesis: size of the analyst’s portfolio (# FIRMS FOLLa,t), analyst following (ANAL FOLLi,t), forecast revision timeliness (QUICK REVi,t), and forecast accuracy (ACCURACYi,t). Similar to the preceding models, the DISTANCEi,t variable is equal to MEAN DIST 10a,t or MEAN # FIRMS 100a,t in the analyst portfolio tests and DIST 10i,t or # FIRMS 100i,t in the analyst following, revision timeliness, and analyst accuracy tests. We proxy for the difficulty in forecasting using the volatility in seasonally differenced quarterly earnings over the prior 3 years.22 We then create an indicator variable equal to one if EARN VOLi,t is above the median and 0 otherwise (HIGH EARN VOLi,t). For our analyst portfolio tests, the partition is based on the mean EARN VOLi,t of the firms in the analysts’ portfolio (HIGH MEAN EARN 21 As in our analysis of analyst following, the univariate correlation between ln(AGEi,t) and ACCURACYi,t is positive (0.09) suggesting the negative coefficient in the regression may be due to the high positive correlation between firm age and other control variables (e.g., firm size, institutional ownership, and volatility). 22 We are attempting to capture forecasting difficulty that is inherent to the firm and not affected by voluntary disclosure choices, such as the provision of guidance. The firm’s geographic distance to other firms in the industry may impact firms’ voluntary disclosure choices. Therefore, we use earnings volatility to measure forecasting difficulty versus past forecast accuracy. 19 VOLa,t). We expect an insignificant coefficient on the DISTANCEi,t variable, suggesting that firms with earnings that are easier to forecast are less likely to be affected by the additional costs associated with being located farther away from other firms in the industry. We expect a negative coefficient on the interaction between the DISTANCEi,t and HIGH EARN VOLi,t variables, suggesting that the firm’s location relative to other firms in the industry is more likely to affect analysts’ cost of gathering and analyzing information when earnings are more difficult to forecast. We also include all previously discussed control variables in each specification. We report the results of estimating Equation 5 in Table 6. Panel A presents the regression results with the size of the analyst’s portfolio (# FIRMS FOLLa,t) as the dependent variable. The coefficient on the MEAN DIST 10a,t variable is equal to 0.000 and insignificant, while the coefficient on the MEAN # FIRMS 100a,t variable is equal to -0.297 and significant at the 1% level. This result provides some evidence that the average distance between the firm and other firms in the industry decreases the number of firms followed by the analyst even when earnings are relatively easy to forecast. More importantly, both coefficients on the interactions between the average distance variables and the difficult to forecast indicator variable are negative and significant at the 1% level, suggesting the effect of distance on analysts’ costs of acquiring and analyzing information is more pronounced when earnings are more difficult to forecast. The sum of the coefficients on the MEAN DIST 10a,t (MEAN # FIRMS 100a,t) variable and the interaction term is equal to -0.298 (-0.558), suggesting that analysts who follow firms that are geographically more distant to other firms in the industry are likely to follow 30% (56%) fewer firms. All control variables have similar signs and magnitudes to those reported in Table 2. In Panel B of Table 6, the dependent variable is analyst following (ln(ANAL FOLLi,t)). The coefficients on DIST 10i,t and # FIRMS 100i,t are equal to -0.019 and 0.012 and are statistically insignificant, suggesting that distance has no effect on analyst following for firms with low earnings volatility. The coefficient on the interaction between the HIGH EARN VOLi,t and DIST 10i,t (# FIRMS 100i,t) variables is -0.088 (-0.087) and significant at the 1% level suggesting that firms with earnings that are more difficult to forecast have lower analyst following when located farther away from other firms in 20 the industry. The sum of the DIST 10i,t (# FIRMS 100i,t) variable and the interaction term is equal to 0.107 (-0.075) and is significant at the 1% level, indicating that firms with high earnings volatility experience a 11% (8%) decrease in analyst following when moving from the lowest to the highest deciles of our distance variables. All other control variables are similar to those reported in Table 3. This evidence suggests that distance is only an important detractor of analyst following when the firm has earnings that are more difficult to forecast, which is consistent with our second hypothesis. Panel C reports the results with analyst forecast timeliness (QUICK REVi,t) as the dependent variable. We find insignificant coefficients of 0.004 and 0.006 on the DIST 10i,t and # FIRMS 100i,t variables, suggesting that distance has no effect on forecast revision timeliness for firms with earnings that are easier to forecast. We find negative and significant (1% level) coefficients of -0.045 and -0.053 on the interaction between the distance variables (DIST 10i,t and # FIRMS 100i,t) and the HIGH EARN VOLi,t variable, suggesting that analyst forecast revisions are less timely for firms with high earnings volatility. The sum of the coefficients on the DIST 10i,t (# FIRMS 100i,t) variable and the interaction term is equal to -0.041 (-0.047) and is significant at the 1% level indicating that firms with high earnings volatility experience a 9.6% (11.0%) decrease in the percentage of analysts revising forecasts (relative to the mean of QUICK REVi,t) on the day of or the day following the annual earnings announcement. These results are also consistent with our second hypothesis. Finally, Panel D reports the results with analyst forecast accuracy (ACCURACYi,t) as the dependent variable. Surprisingly, we find positive and significant (1% level) coefficients on the DIST 10i,t and # FIRMS 100i,t variables of 0.224 and 0.205, respectively. These results suggest that distance improves analyst forecasting ability for firms with low earnings volatility, which is contrary to expectations. However, for firms with high earnings volatility, the effect reverses. The coefficient on the interaction between the HIGH EARN VOLi,t and DIST 10i,t (# FIRMS 100i,t) variables is -0.347 (-0.293) and significant at the 1% level, suggesting that firms with earnings that are more difficult to forecast have lower forecast accuracy when located farther away from other firms in the industry. The sum of the coefficients on the DIST 10i,t (# FIRMS 100i,t) variable and the interaction term is equal to -0.123 (-0.088) 21 and is significant at the 5% (10%) level, indicating that for firms with high earnings volatility moving from the bottom to the top decile of the distance measure reduces analyst forecast accuracy by 19.5% (13.9%), relative to the mean of ACCURACYi,t. These results contrast those reported in Table 5 and suggest distance reduces forecast accuracy only for firms with earnings that are more difficult to forecast, which is consistent with our second hypothesis. Overall, we find largely consistent evidence that distance increases analysts’ gathering and analyzing costs, primarily for firms with earnings that are more difficult to forecast. IV. ANALYSIS OF MANAGEMENT COMMUNICATION The prior evidence is consistent with our hypothesis that geographic proximity to other firms in the industry produces information spillover effects that enhance managers’ knowledge of other firms in the area. This enhanced knowledge reduces analysts’ information acquisition costs, assuming managers communicate their knowledge to analysts. To provide some evidence on this issue, we examine the likelihood that a manager references geographically close firms in their industry when communicating with analysts, investors, and other market participants. Specifically, we examine whether firms are more likely to mention, during their earnings conference calls, other firms in the same industry as the distance between them decreases. We recognize that conference calls are just one mechanism through which managers might convey their enhanced knowledge of other firms in their industry to analysts. As discussed previously, informal meetings that occur during site visits as well as during one-on-one phone calls also represent avenues through which managers can communicate their knowledge. However, these modes of communication are not directly observable. For our sample firms, we obtain conference call manuscripts from Factiva’s FD Wire from 2002 to 2011. We begin our analysis in 2002 because this is when Factiva began collecting conference calls. We find 41,365 total conference calls for 13,251 of our sample firms-years. The mean (median) firm has 3.12 (3) calls each year, which is reasonable since firms typically have one conference call per quarter. For each firm-year, we obtain the names of all other firms in the same industry from Compustat and 22 search the conference call transcripts for mentions of these firms’ names.23 We then create a dataset of all firm i-j matches, where firm i is the firm holding the conference call and firm j is a firm in the same industry as firm i. We do not require that firm j hold a conference call during the year. The resulting dataset consists of 4.54 million unique observations. We then create an indicator variable (MENTIONi,j,t) that is equal to one if the managers of firm i or the analysts involved in the conference call mention firm j in any of firm i’s conference calls during year t. We also create two additional indicator variables equal to one if the managers of the firm mention the industry firm first (MENTION (FIRM FIRST)i,j,t) and if the analysts mention the industry firm first (MENTION (ANAL FIRST)i,j,t). In these tests, we do not use the DIST 10i,t and # FIRM 100i,t variables (as previously described) because we are looking at unique firm i-j combinations rather than average firm i-j combinations for all firms in the industry. The DISTANCEi,j,t variable is our variable of interest and represents the decile ranked distance between the headquarters of firm i and firm j during year t.24 We find 16,594 total mentions of industry firms in our sample. While this may appear small relative to the total number of paired observations in our sample, we find that 46% of our sample firms mention at least one other firm in the same industry in their conference calls during the year. Based on the significant differences in firm specific characteristics discussed in Section 3, we test whether firms are less likely to mention more geographically distant industry firms by estimating the following logistic regression. CONF CALL MENTIONSi,j,t = α0 + α1 DISTANCEi,j,t + α2 ln(MKT VALi,t) + α3 ln(MKT VALj,t) + α4 (6) BTMi,t + α5 BTMj,t + α6 INST OWNi,t + α7 INST OWNj,t + α8 ROAi,t + α9 ROAj,t + α10 TURNi,t + α11 TURNj,t + α12 BUS CONCi,t + α13 23 A potential concern is the use of abbreviations such as “corp” instead of “corporation” or simply omitting the word “corporation” when referencing another firm in the industry. We attempt to mitigate this problem by removing all instances of the words “corporation” or “incorporated” (and their counterparts “corp” and “inc”) in our search. For firms in which the company name is ambiguous, we include the word “corp” or “inc” in our search. For example, we include “corp” for “News Corp” but exclude “corp” for “Microsoft Corp.” We also manually examine each company name to identify potential alternative abbreviations likely used by analysts (e.g., Exxon for ExxonMobile). 24 The ranking process of the DISTANCE i,j,t variable is similar to that for the DIST 10i,t and # FIRMS 100i,t variables. Using all firm combinations, we decile rank the distance between firm i and j, subtract 1 and divide by 9 to obtain a variable ranking taking on values between 0 and 1. 23 BUS CONCj,t + α14 EARN VOLi,t + α15 EARN VOLj,t + α16 RET VOLi,t + α17 RET VOLj,t + + α18 ln(AGEi,t) + α19 ln(AGEj,t) + α20 ln(ANAL FOLLi,t) + α21 ln(ANAL FOLLj,t) + α22 COMPETITORi,j,t + α23 EARN COMOVEi,j,t + YEAR + INDUSTRY + εi,t CONF CALL MENTIONSi,j,t equals either MENTIONSi,j,t, MENTIONS(FIRM FIRST)i,j,t, or MENTIONS(ANAL FIRST)i,j,t. We expect a negative coefficient on the DISTANCEi,j,t variable. We include control variables for both the mentioning firm (firm i) and the mentioned firm (firm j) since the firm characteristics of either firm could influence the decision to mention a firm. We also include an indicator variable equal to 1 if firm i and firm j are competitors and 0 otherwise (COMPETITORi,j,t). We define competitors using the Hoberg and Phillips (2010) text-based network industry definitions, which use contextual analysis of the product descriptions in the 10-K’s to identify product competitors. This variable represents a potentially correlated omitted variable if firms locate closer to competitors and if firms are more likely to mention competitors during conference calls. We also include the co-movement of earnings between firm i and firm j (EARN COMOVEi,j,t), defined as the correlation between the quarterly operating incomes scaled by lagged total assets of firm i and firm j over the twelve quarters prior to period t, to control for the possibility that more similar firms (i.e., those that respond similarly to industry shocks) locate more closely together and are more likely to mention each other in their conference calls. We also include year and industry (two-digit SIC code) indicator variables to control for differences in mentions over time and across industries. Finally, we cluster the standard errors by firm i-j combinations due to potential serial correlation in our dependent and independent variables (Petersen, 2009). We report the results of estimating Equation 6 in Table 7. We note that the coefficient on the DISTANCEi,j,t variable is equal to -0.346 in Column 1, suggesting that managers and analysts are less likely to mention other firms in the same industry when those firms are geographically farther away. We also note that the odds ratio is equal to 0.71, suggesting that management and analysts are 29% less likely to mention a firm in the same industry that is in the 10th distance decile compared to a firm in the 1st distance decile. These results suggest that managers communicate their knowledge and awareness of other 24 industry firms in the same geographic area with analysts and investors during the earnings conference call.25 We find similar results in Columns 2 and 3 when the managers of the firm are the first to mention the industry firm and when the analysts are the first to mention the industry firm, respectively. Specifically, the coefficient on DISTANCEi,j,t in Column 2 (Column 3) is equal to -0.308 (-0.409) and the odds ratio is 0.73 (0.66). These results suggest that both managers and analysts are less likely to mention industry firms that are located farther away from the firm. We find that both characteristics of the mentioning firm as well as the mentioned firm influence the decision to mention another firm. In particular, we note that firms are more likely to mention direct competitors. We also note that managers are more likely to mention other industry firms that are older, larger, have higher return volatility, have greater earnings co-movement, and have greater analyst following. Interestingly, we find that higher institutional ownership and higher ROA reduces the likelihood that managers mention another firm in the industry. V. ROBUSTNESS TESTS Changes Analysis Our primary analysis is based on a “levels” specification because changes in firms’ headquarter locations are relatively rare events. Moreover, the decision to move locations is not likely exogenous; thus, the benefit of a changes specification relative to the costs associated with a much smaller sample size is likely low. Nevertheless, we examine the effect of headquarter relocations on changes in analyst characteristics surrounding these relocation events. We expect analysts’ information acquisition costs to decrease when the firm relocates closer to other firms in the industry and to increase when the firm moves farther away. We identify relocating firms as those in which the headquarters location in year t is greater 25 We also run an alternative specification at the firm/year level. The dependent variable is equal to one if firm i mentions another firm in the same industry during year t, and our variable of interest is either DIST 10 or # FIRMS 100. The interpretation of these results, however, is somewhat different. A significant coefficient on DIST 10 (#FIRMS 100) indicates that firms are more likely to mention another firm in their industry when they are located closer to other firms in the industry (irrespective of the distance of the specific firm mentioned). Results suggest this is the case, which is consistent with the results of our analysis reported here. 25 than or equal to 200 miles from the headquarters location in year t-1. We choose 200 miles to ensure the firm is truly relocating away from industry firms in its previous geographical area. We also eliminate observations where the move appears to be the result of a merger. We find a total of 192 total relocations for our sample firms. Using a multivariate analysis, we examine whether the changes (from year t-1 to t+1) in analyst following, timeliness of revisions, and forecast accuracy decrease more for firms that move farther away from other industry firms compared to firms who move closer. We identify firms that move farther away by examining whether the change in DIST 10i,t and #FIRMS 100i,t increase from year t-1 to t+1. We also include the changes in control variables from year t-1 to t+1 because changes in firm characteristics can influence a firm’s decision to re-locate. The control variables include changes in market value, book to market ratio, institutional ownership, return on assets, stock turnover, business concentration, earnings volatility, returns volatility, and earnings co-movement within the industry. Finally, we include year and industry fixed effects as in prior tests. Using the change in the # FIRMS 100i,t variable, we find that firms that move farther away relative to those firms that move closer to industry peers have lower analyst following, which is marginally significant at the 10% level. We also find evidence that analysts’ forecast accuracy decreases at the 1% and 10% significance level using the DIST 10i,t and #FIRMS 100i,t variables for firms that move farther away compared to firms who move closer to industry peers. We do not find any evidence that the percentage of analyst who revise their forecasts in a timely manner decreases more for firms that move farther away relative to those firms who move closer to industry peers. Given the relatively small sample size, we view the evidence based on this changes analysis as supportive of our prior results, suggesting that proximity to other firms in the industry reduces analysts’ information acquisition costs. Alternative Measurements and Controls We also perform several robustness tests to examine the sensitivity of our analyses to various measurement choices and additional control variables. First, we test the sensitivity of our industry 26 definitions. We use two-digit SIC codes throughout the main body of the paper because we want to use a definition of industry that includes not only a firm’s direct competitors but also firms in the supply chain (because information spillovers can occur across suppliers and customers as well). However, we also reperform each test using the Hoberg and Phillips (2010) industry definitions to identify firm competitors.26 Using the Hoberg and Phillips (2010) industry classifications, we find similar results (untabulated) for our main analyses (Tables 3, 4, and 5). For the cross-sectional analysis based on forecasting difficulty (Table 6), we find similar results with the following exceptions: 1) analyst following is lower when the firm is farther away from other firms within the industry regardless of whether earnings are more or less difficult to forecast (i.e., the main effect of DIST 10i,t # FIRMS 100i,t is significantly negative but the interaction terms with HIGH EARN VOLit are no longer significantly negative) and 2) in the forecast revision timeliness regression, the interaction between HIGH_EARN_VOLit and #FIRMS100i,t is not significant at conventional levels (t-stat of 1.60).. These tests suggest that our main results are not solely driven by the co-location of significant suppliers and customers, who are more likely to share information because of their shared operations, but also by competitors who locate in close proximity to the firm. Next, our distance measures are potentially influenced by the size of the industry. For example, a firm operating in an industry with 20 firms is less likely to have firms nearby than a firm operating in an industry with 200 firms. In our main analysis, we control for this problem by including industry fixed effects so that our distance measures are capturing a within-industry effect. However, we directly address this issue using two additional sensitivity tests. First, we find qualitatively similar results (untabulated) when we specifically control for the number of firms in the industry for each year. Second, we re-define our distance measures to be relative to the size of the firm’s industry. We convert our two distance variables into percentage variables: 1) DIST 10%i,t, defined as the distance between the firm and the closest 10% of firms in the industry and 2) %FIRMS 100i,t, defined as the percentage of firms in the industry within 100 miles of the firm. Our results (untabulated) are generally consistent with these 26 Hoberg and Phillips (2010) use product descriptions from mandatory disclosures to identify firms with the most similar activities to identify text-based network industry classifications. 27 alternative definitions with three exceptions. First, we do not find significant coefficients on DIST 10%i,t and %FIRMS 100i,t in our main analyst following tests (Table 3) However, if we instead define distance as the percentage of firms within 50 miles of the firm (%FIRMS 50i,t), then we find a significantly negative coefficient on the distance variable at the 5% level. Second, we do not find significantly negative coefficients on either the main or interaction coefficients in the cross-sectional analyst following tests (Table 6 Panel A) using %FIRMS 100i,t as our measure of distance; however, we do find a significantly negative interaction coefficient on DIST 10%i,t and HIGH EARN VOLi,t, at the 10 percent level. Third, the DIST 10%i,t variable is insignificant in the main analyst portfolio test (Table 2), but its interaction with high earnings volatility is negative and significant in the cross-sectional test (Table 6, Panel A). In general, our results, particularly with respect to analyst following, are somewhat sensitive to defining distance based on the percentage of firms nearby versus defining distance based on the number of firms nearby. However, because our results are robust to both including industry fixed effects as well as to including the number of firms in the industry as an additional control variable, we do not believe the sensitivity of our results to the use of percentages is the result of the size of the industry. Rather, we interpret these results as evidence that it is the number of the firms in the industry rather than the percentage of the firms in the area that influence analysts’ information acquisition costs. Next, Loughran and Schulz (2005) examine whether firms in rural areas are followed by fewer analysts than firms in urban areas. In untabulated analysis, we verify that our distance measures are not merely capturing a “rural” or “urban” effect by including indicator variables for rural and urban firms in the analyst following model. Specifically, following Loughran and Schulz (2005), we include a rural indicator variable equal to 1 if the firm is located 100 miles or more from the center of any of the 49 U.S. metropolitan statistical areas with one million or more people. We also include an urban indicator variable equal to 1 if the firm is located in one of the ten most populated metropolitan statistical areas during year t: New York, Los Angeles, Chicago, Washington-Baltimore, San Francisco, Dallas, Philadelphia, Boston, Houston, or Detroit. The sign and significance of the coefficients on the DISTANCEi,t variables are unaffected by the inclusion of the rural and urban indicator variables. 28 Controlling for the mean effect of being located in an urban area also addresses the concern that our results reflect the distance of the firm to the sell-side analyst. Malloy (2005) provides evidence that sellside analysts who are closer to the firm are likely to have more accurate recommendations. To the extent that firms located in urban areas are geographically closer to a larger proportion of the analysts covering the firm, our distance variable could be capturing the distance of the firm to the analyst (versus the information-spillover effect associated with being closer to other firms in the industry). The inclusion of the urban indicator variable addresses this concern.27 VI. CONCLUSION We provide evidence that the co-location of firms within an industry affects analysts’ costs of gathering and processing information. Based on the prior research, we argue that firm managers who are located closer to other firms in the same industry and geographic area are more likely to know more about these firms, increasing the level of industry specific information held by these managers. If managers communicate this information to analysts, we expect analysts’ costs of gathering and processing information to decrease as they follow firms that are located geographically closer to other firms in the industry. We provide evidence consistent with this prediction. First, we find that analyst portfolios contain fewer firms when the firms included in the portfolio are located farther away from other industry firms. We also find that fewer analysts follow firms that are located farther away from other firms in the industry and they are less likely to revise their forecasts immediately following the earnings announcement, suggesting increased costs to follow these firms. In additional cross-sectional tests, we find that the increased analyst costs that are associated with being farther away from other firms in the same industry are less of a concern when earnings are easier to forecast: the negative impact of distance on the size of analysts’ portfolios, the number of analysts following a firm, the timeliness of forecast revisions, the accuracy of analysts’ forecasts is greater when earnings are more difficult to forecast. 27 Directly controlling for the average distance between the firm and the analysts covering the firm is problematic because the location of the analyst is not readily measurable. 29 Lastly, we provide supporting evidence that managers are more knowledgeable about firms in the same industry that are geographically closer by examining the location of those firms mentioned by management during conference calls. We find that firms are less likely to mention other firms in the industry during conference calls as they become more geographically distant, providing some evidence that firms are more knowledgeable about firms that are geographically closer. To the best of our knowledge, we are the first to thoroughly examine how the firm’s geographic location affects analysts’ information acquisition costs. Our results are of potential interest to investors in understanding how the location of the firm relative to other firms in the industry affects information intermediaries’ costs of obtaining information about the firm. In this paper, we do not address the specific mechanisms by which the information is transferred. We leave this research question open to future research. 30 REFERENCES Alcácer, J. 2006. Location Choices Across the Value Chain: How Activity and Capability Influence Collocation. Management Science 52 (10): 1457–1471. Alford, A. W., and P. G. Berger. 1999. A Simultaneous Equations Analysis of Forecast Accuracy, Analyst Following, and Trading Volume. Journal of Accounting, Auditing & Finance 14 (3): 219–240. Asquith, P., M. B. Mikhail, and A. S. Au. 2005. Information content of equity analyst reports. Journal of Financial Economics 75 (2): 245–282. Audretsch, D. B., and M. P. Feldman. 1996. R&D Spillovers and the Geography of Innovation and Production. The American Economic Review 86 (3): 630–640. Bae, K.-H., R. M. Stulz, and H. Tan. 2008. Do local analysts know more? A cross-country study of the performance of local analysts and foreign analysts. Journal of Financial Economics 88 (3): 581– 606. Baum, J. A. C., and H. A. Haveman. 1997. Love Thy Neighbor? Differentiation and Agglomeration in the Manhattan Hotel Industry, 1898-1990. Administrative Science Quarterly 42 (2): 304–338. Beyer, A., D. A. Cohen, T. Z. Lys, and B. R. Walther. 2010. The financial reporting environment: Review of the recent literature. Journal of Accounting and Economics 50 (2–3): 296–343. Bhushan, R. 1989. Firm characteristics and analyst following. Journal of Accounting and Economics 11 (2–3): 255–274. Brown, L. D., R. L. Hagerman, P. A. Griffin, and M. E. Zmijewski. 1987. An evaluation of alternative proxies for the market’s assessment of unexpected earnings. Journal of Accounting and Economics 9 (2): 159–193. Brown, L. D., and M. S. Rozeff. 1978. The Superiority of Analyst Forecasts as Measures of Expectations: Evidence from Earnings. The Journal of Finance 33 (1): 1–16. Chhaochharia, V., A. Kumar, and A. Niessen-Ruenzi. 2012. Local investors and corporate governance. Journal of Accounting and Economics 54 (1): 42–67. Coval, J. D., and T. J. Moskowitz. 1999. Home Bias at Home: Local Equity Preference in Domestic Portfolios. The Journal of Finance 54 (6): 2045–2073. Engelberg, J., A. Ozoguz, and S. Wang. 2010. Know Thy Neighbor: Industry Clusters, Information Spillovers and Market Efficiency. Working paper. Frankel, R., S. P. Kothari, and J. Weber. 2006. Determinants of the informativeness of analyst research. Journal of Accounting and Economics 41 (1–2): 29–54. Fried, D., and D. Givoly. 1982. Financial analysts’ forecasts of earnings: A better surrogate for market expectations. Journal of Accounting and Economics 4 (2): 85–107. Gilson, S. C., P. M. Healy, C. F. Noe, and K. G. Palepu. 2001. Analyst Specialization and Conglomerate Stock Breakups. Journal of Accounting Research 39 (3): 565–582. Givoly, D., and J. Lakonishok. 1979. The information content of financial analysts’ forecasts of earnings: Some evidence on semi-strong inefficiency. Journal of Accounting and Economics 1 (3): 165– 185. Green, T. C., R. Jame, S. Markov, and M. Subasi. 2012. Access to Management and the Informativeness of Analyst Research. Working paper. Hayes, R. M. 1998. The Impact of Trading Commission Incentives on Analysts’ Stock Coverage Decisions and Earnings Forecasts. Journal of Accounting Research 36 (2): 299–320. Herbalife LTD. January 10, 2013. Analyst and investor meeting transcript. Hoberg, G., and G. Phillips. 2010. Product Market Synergies and Competition in Mergers and Acquisitions: A Text-Based Analysis. Review of Financial Studies 23 (10): 3773–3811. Ivković, Z., and S. Weisbenner. 2005. Local Does as Local Is: Information Content of the Geography of Individual Investors’ Common Stock Investments. The Journal of Finance 60 (1): 267–306. Jaffe, A. B., M. Trajtenberg, and R. Henderson. 1993. Geographic Localization of Knowledge Spillovers as Evidenced by Patent Citations. The Quarterly Journal of Economics 108 (3): 577–598. 31 Lang, M. H., and R. J. Lundholm. 1996. Corporate Disclosure Policy and Analyst Behavior. The Accounting Review 71 (4): 467–492. Loughran, T., and P. Schultz. 2004. Weather, Stock Returns, and the Impact of Localized Trading Behavior. The Journal of Financial and Quantitative Analysis 39 (2): 343–364. ———. 2005. Liquidity: Urban versus rural firms. Journal of Financial Economics 78 (2): 341–374. Lovely, M. E., S. S. Rosenthal, and S. Sharma. 2005. Information, agglomeration, and the headquarters of U.S. exporters. Regional Science and Urban Economics 35 (2): 167–191. Malloy, C. J. 2005. The Geography of Equity Analysis. The Journal of Finance 60 (2): 719–755. Marshall, A. 1920. Principles of Economics. 8th ed. London: Macmillan. Orpurt, S. 2004. Local Analyst Earnings Forecast Advantages in Europe. Working paper. Petersen, M. A. 2009. Estimating Standard Errors in Finance Panel Data Sets: Comparing Approaches. Review of Financial Studies 22 (1): 435–480. Shaver, Myles, J., and F. Flyer. 2000. Agglomeration economies, firm heterogeneity, and foreign direct investment in the United States. Strategic Management Journal 21 (12): 1175–1193. Stein, J. C. 2008. Conversations among Competitors. American Economic Review 98 (5): 2150–2162. Thomas, S. 2002. Firm diversification and asymmetric information: evidence from analysts’ forecasts and earnings announcements. Journal of Financial Economics 64 (3): 373–396. Zhang, Y. 2008. Analyst responsiveness and the post-earnings-announcement drift. Journal of Accounting and Economics 46 (1): 201–215. Zhu, N. 2002. The Local Bias of Individual Investors. Unpublished working paper. 32 APPENDIX A: VARIABLE DEFINITIONS Variable # FIRMS 100i,t Definition The decile rank (between 0 and 9 and divided by 9) of the number of firm within 100 miles of firm i's headquarters during year t multiplied by negative one # FIRMS FOLLa,t The number of firms followed by analyst a during year t. # FIRMS FOLLi,t The average # FIRMS FOLLa,t for all analysts following firm i in year t. # IND FIRMSi,t # INDUSTRIESa,t The number of firms in firm i's industry during year t. The number of industries followed by analyst a during year t. # INDUSTRIESi,t The average # INDUSTRIESa,t for all analysts following firm i in year t. ACCURACYi,t The average over the four quarters during year t of the absolute value of the quarterly analyst consensus forecast less actual EPS divided by price multiplied by -1 for firm i in year t. AGEi,t The number of years between the fiscal period end date and the date the firm first appears in the Compustat database for firm i in year t. ANAL FOLLi,t The number of analysts following firm i in year t. BROKER SIZEa,t The normalized rank of analyst a's brokerage house based on the number of firms followed by the broker during year t. BROKER SIZEi,t The average BROKER SIZEa,t for all analysts following firm i in year t. BTMi,t The book value of equity divided by the market value of equity for firm i in year t. BUS CONCi,t The firm's business segment concentration defined as the firmspecific Herfindahl-Hirschman Index computed for firm i's industry business segment sales in year t. COMPETITORi,j,t An indicator variable equal to 1 if firm j is one of firm i's competitors in year t based on the Hoberg and Phillips (2010) text-based network industry classifications. DIST 10i,t The decile rank (between 0 and 9 and divided by 9) of the mean number of miles from firm i to its 10 geographically closest industry peers during year t. DISTANCEi,j,t The decile rank (between 0 and 9 and divided by 9) of the number of miles between firm i and firm j in year t. 33 EARN COMOVEi,t EARN VOLi,t The average correlation between firm i’s operating income scaled by lagged total assets and the operating incomes scaled by lagged total assets of the ten geographically closest firms in the same industry over the previous 12 quarters (requiring a minimum of 8 prior quarters to compute this variable) The standard deviation of seasonally-differenced earnings over the last 12 quarters for firm i at the end of year t (requiring a minimum of 8 prior quarters to compute this variable) EXPERIENCEa,t The number of years since analyst a first appears in the IBES database as of year t. EXPERIENCEi,t The average EXPERIENCEa,t for all analysts following firm i in year t. HIGH EARN VOLi,t An indicator variable equal to 1 if EARN VOLi,t is above the median and 0 otherwise. INST OWNi,t The percentage of firm i's shares held by institutional investors in year t. MEAN # FIRMS 100a,t The mean of # FIRM 100i,t for all firms followed by analyst a during year t. MEAN BTMa,t The mean of BTMi,t for all firms followed by analyst a during year t. MEAN BUS CONCa,t The mean of BUS CONCi,t for all firms followed by analyst a during year t. MEAN DIST 10a,t MEAN EARN COMOVEa,t The mean of DIST 10i,t for all firms followed by analyst a during year t. The mean of EARN COMOVEi,t for all firms followed by analyst a during year t. MEAN EARN VOLa,t The mean of EARN VOLi,t for all firms followed by analyst a during year t. MEAN INST OWNa,t The mean of INST OWNi,t for all firms followed by analyst a during year t. MEAN ln(AGE)a,t MEAN ln(MKT VAL)a,t MEAN RET VOLa,t The mean of ln(AGEi,t) for all firms followed by analyst a during year t. The mean of ln(MKT VALi,t) for all firms followed by analyst a during year t. The mean of RET VOLi,t for all firms followed by analyst a during year t. 34 MEAN ROAa,t The mean of ROAi,t for all firms followed by analyst a during year t. MEAN TURNa,t The mean of TURNi,t for all firms followed by analyst a during year t. MENTIONi,j,t An indicator variable equal to 1 if firm i mentions firm j (a firm in the same industry as firm i) in any of its conference calls during year t. MKT VALi,t QUICK REVi,t The market value of equity of firm i at the end of year t. The average over the four quarters during year t of the percentage of analyst forecasts revised on day 0 or day 1 following the quarterly earnings announcements for firm i in year t. RET VOLi,t The standard deviation of monthly returns over the last 12 months for firm i at the end of year t. ROAi,t The net income before extraordinary items divided by lagged total assets for firm i in year t. TURNi,t The percentage of firm i's shares traded during year t. 35 TABLE 1 Descriptive Statistics This table presents the means of firm-specific variables for the bottom and top halves of the DIST 10i,t variable and for the full sample. We also report the p-value of the t-test of the difference in means between the two groups. *, **, and *** represent significance at the 10%, 5%, and 1% levels, respectively. All variables are defined in Appendix A. All continuous variables are winsorized at the 1st and 99th percentiles. DIST 10i,t Partition Bottom N = 16,510 Top N = 16,511 All N = 33,021 Top v. Bottom DIST 10i,t (UNRANKED) 10.933 232.461 121.707 0.00*** # FIRMS 100i,t (UNRANKED) -54.168 -3.958 -29.061 0.00*** MKT VALi,t 3,447.3 2,854.8 3,151.0 0.10 BTMi,t 0.460 0.579 0.520 0.00*** INST OWNi,t 0.589 0.600 0.595 0.11 ROAi,t -0.012 0.045 0.016 0.00*** TURNi,t 0.023 0.019 0.021 0.00*** BUS CONCi,t 0.939 0.903 0.921 0.00*** EARN VOLi,t 0.047 0.026 0.036 0.00*** RET VOLi,t 0.151 0.127 0.139 0.00*** AGEi,t 17.101 20.961 19.031 0.00*** EARN COMOVEi,t 0.061 0.071 0.066 0.00*** Variable 36 TABLE 2 Number of Firms Followed Regression Results. This table includes all analyst/year observations from 1994 to 2011 with sufficient data to calculate the dependent and independent variables. The dependent variable is the natural logarithm of the number of firms followed by analyst a during year t (ln(# FIRMS FOLLa,t)). Columns 1 and 2 include the mean of DIST 10i,t and # FIRMS 100i,t, respectively, for all firms followed by analyst a during year t as the independent variables of interest (MEAN DIST 10a,t and MEAN # FIRMS 100a,t). All variables are defined in Appendix A. Year fixed effects are included as additional independent variables. The coefficients on the year indicator variables are suppressed. Standard errors are clustered by analyst. All continuous variables are winsorized at the 1% and 99% levels. *, **, and *** represent significance at the 10%, 5%, and 1% levels, respectively. INTERCEPT MEAN DIST 10a,t [1] ln(# FIRMS FOLLa,t) [2] ln(# FIRMS FOLLa,t) -0.565*** (-7.007) -0.191*** (-11.213) -0.272*** (-3.391) 0.976*** (29.647) 0.372*** (68.294) 0.743*** (66.483) 0.019*** (3.847) 0.000 (1.374) -0.361*** (-8.982) -0.001 (-1.314) 0.163 (0.494) 1.069*** (18.404) 0.017 (1.311) -0.716*** (-6.731) -0.021 (-1.546) 0.402*** (10.396) -0.455*** (-27.073) 0.978*** (30.350) 0.363*** (67.527) 0.737*** (68.378) 0.004 (0.895) 0.000 (1.179) -0.351*** (-8.947) -0.001 (-1.278) -0.322 (-0.994) 0.954*** (16.520) 0.008 (0.851) -1.023*** (-9.953) 0.014 (1.041) 0.468*** (12.437) 56,128 0.459 56,128 0.478 MEAN # FIRMS 100a,t BROKER SIZEa,t ln(EXPERIENCEa,t) ln(# INDUSTRIESa,t) MEAN ln(MKT VAL)a,t MEAN BTMa,t MEAN INST OWNa,t MEAN ROAa,t MEAN TURNa,t MEAN BUS CONCa,t MEAN EARN VOLa,t MEAN RET VOLa,t MEAN ln(AGE)a,t MEAN EARN COMOVEa,t #OBS Adjusted R2 37 TABLE 3 Analyst Following Regression Results This table includes all firm/year observations from 1994 to 2011 with sufficient data to calculate the dependent and independent variables. The dependent variable is the natural logarithm of the number of analysts following firm i during year t (ln(ANAL FOLLi,t)). Column 1 (2) includes DIST 10i,t (# FIRMS 100i,t) as the independent variable of interest. All variables are defined in Appendix A. Year and industry (two-digit SIC code) fixed effects are included as additional independent variables. The coefficients on the year and industry indicator variables are suppressed. Standard errors are clustered by firm. All continuous variables are winsorized at the 1% and 99% levels. *, **, and *** represent significance at the 10%, 5%, and 1% levels, respectively. INTERCEPT [1] [2] ln(ANAL FOLLi,t) ln(ANAL FOLLi,t) 0.091 (0.789) 0.054 (0.470) -0.075*** (-3.318) DIST 10i,t -0.042* (-1.896) # FIRMS 100i,t ln(MKT VALi,t) 0.332*** (87.875) 0.333*** (88.076) BTMi,t 0.159*** (13.454) 0.160*** (13.526) INST OWNi,t 0.347*** (15.493) 0.347*** (15.475) ROAi,t -0.398*** (-14.586) -0.403*** (-14.769) TURNi,t 7.599*** (28.181) 7.628*** (28.262) BUS CONCi,t 0.155*** (5.281) 0.159*** (5.400) EARN VOLi,t -0.250*** (-3.145) -0.240*** (-3.012) RET VOLi,t -0.260*** (-4.427) -0.256*** (-4.364) ln(AGEi,t) -0.108*** (-13.697) -0.108*** (-13.821) EARN COMOVEi,t 0.058*** (3.300) 0.060*** (3.405) 33,021 33,021 0.673 0.673 #OBS 2 Adjusted R 38 TABLE 4 Analyst Forecast Revision Timeliness Regression Results This table includes all firm/year observations from 1994 to 2011 with sufficient data to calculate the dependent and independent variables. The dependent variable is the average percentage of analyst forecasts revised on day 0 or day 1 following the quarterly earnings announcement for firm i in year t (QUICK REVi,t). Column 1 (2) includes DIST 10i,t (# FIRMS 100i,t) as the independent variable of interest. All variables are defined in Appendix A. Year and industry (two-digit SIC code) fixed effects are included as additional independent variables. The coefficients on the year and industry indicator variables are suppressed. Standard errors are clustered by firm. All continuous variables are winsorized at the 1% and 99% levels. *, **, and *** represent significance at the 10%, 5%, and 1% levels, respectively. INTERCEPT DIST 10i,t [1] QUICK REVi,t [2] QUICK REVi,t -0.567*** (-12.342) -0.022** (-2.462) -0.568*** (-12.399) -0.002 (-0.976) 0.004 (0.699) 0.071*** (7.035) -0.020* (-1.687) 0.160 (1.393) 0.054*** (4.161) 0.044 (1.272) -0.050* (-1.840) -0.017*** (-5.261) 0.008 (0.916) 0.063*** (15.366) 0.416*** (17.593) 0.006 (0.780) 0.027*** (5.611) -0.023** (-2.534) -0.002 (-0.949) 0.004 (0.725) 0.071*** (7.020) -0.020* (-1.734) 0.157 (1.366) 0.055*** (4.205) 0.045 (1.279) -0.050* (-1.843) -0.017*** (-5.314) 0.008 (0.961) 0.063*** (15.426) 0.415*** (17.524) 0.006 (0.789) 0.027*** (5.631) 33,021 0.521 33,021 0.521 # FIRMS 100i,t ln(MKT VALi,t) BTMi,t INST OWNi,t ROAi,t TURNi,t BUS CONCi,t EARN VOLi,t RET VOLi,t ln(AGEi,t) EARN COMOVEi,t ln(ANAL FOLLi,t) BROKER SIZEi,t ln(# FIRMS FOLLi,t) ln(EXPERIENCEi,t) #OBS Adjusted R2 39 TABLE 5 Analyst Forecast Accuracy Regression Results This table includes all firm/year observations from 1994 to 2011 with sufficient data to calculate the dependent and independent variables. The dependent variable is average quarterly analyst forecast accuracy (ACCURACYi,t) for firm i in year t. Column 1 (2) includes DIST 10i,t (# FIRMS 100i,t) as the independent variable of interest. All variables are defined in Appendix A. Year and industry (two-digit SIC code) fixed effects are included as additional independent variables. The coefficients on the year and industry indicator variables are suppressed. Standard errors are clustered by firm. All continuous variables are winsorized at the 1% and 99% levels. *, **, and *** represent significance at the 10%, 5%, and 1% levels, respectively. INTERCEPT DIST 10i,t [1] ACCURACYi,t [2] ACCURACYi,t -0.575*** (-2.986) 0.028 (0.683) -0.590*** (-3.088) 0.056*** (4.936) -0.196*** (-5.373) 0.443*** (9.739) 0.742*** (9.397) 1.725** (2.478) -0.056 (-1.195) -3.333*** (-12.880) -3.944*** (-18.661) -0.136*** (-9.365) -0.005 (-0.118) 0.209*** (10.474) 0.168 (1.273) -0.001 (-0.029) -0.042* (-1.734) 0.046 (1.232) 0.056*** (4.944) -0.196*** (-5.379) 0.443*** (9.743) 0.741*** (9.409) 1.743** (2.504) -0.056 (-1.195) -3.329*** (-12.870) -3.942*** (-18.664) -0.136*** (-9.346) -0.005 (-0.123) 0.209*** (10.489) 0.171 (1.293) -0.001 (-0.025) -0.042* (-1.748) 33,021 0.276 33,021 0.276 # FIRMS 100i,t ln(MKT VALi,t) BTMi,t INST OWNi,t ROAi,t TURNi,t BUS CONCi,t EARN VOLi,t RET VOLi,t ln(AGEi,t) EARN COMOVEi,t ln(ANAL FOLLi,t) BROKER SIZEi,t ln(# FIRMS FOLLi,t) ln(EXPERIENCEi,t) #OBS Adjusted R2 40 TABLE 6 Cross Sectional Regression Results for Analyst Properties This table includes all firm/year observations from 1994 to 2011 with sufficient data to calculate the dependent and independent variables. The dependent variables in Panels A through D are ln(# FIRMS FOLLa,t), ln(ANAL FOLLi,t), QUICK REVi,t, and ACCURACYi,t, respectively. All variables are defined in Appendix A. Year fixed effects are included as additional independent variables. Industry (two-digit SIC code) fixed effects are included in Panels B through D. The coefficients on the year and industry indicator variables are suppressed. Standard errors are clustered by firm. All continuous variables are winsorized at the 1% and 99% levels. *, **, and *** represent significance at the 10%, 5%, and 1% levels, respectively. Panel A. Number of Firms Followed by Analysts INTERCEPT MEAN DIST 10a,t MEAN DIST 10a,t * HIGH MEAN EARN VOLa,t [1] ln(# FIRMS FOLLa,t) [2] ln(# FIRMS FOLLa,t) -0.661*** (-8.105) 0.000 (0.016) -0.298*** (-10.242) -0.354*** (-4.340) MEAN # FIRMS 100a,t 0.268*** (12.995) 0.977*** (29.925) 0.368*** (67.216) 0.737*** (66.369) 0.024*** (4.711) 0.000 (1.389) -0.374*** (-9.239) -0.000 (-0.762) -0.634* (-1.947) 1.042*** (17.885) -1.077*** (-10.519) -0.017 (-1.244) 0.405*** (10.649) -0.297*** (-11.692) -0.261*** (-8.772) 0.187*** (9.306) 0.972*** (30.252) 0.361*** (67.011) 0.739*** (68.916) 0.007 (1.456) 0.000 (1.179) -0.364*** (-9.231) -0.000 (-0.759) -0.788** (-2.424) 0.939*** (16.225) -1.197*** (-11.998) 0.015 (1.117) 0.473*** (12.675) 56,128 0.464 56,128 0.481 MEAN # FIRMS 100a,t * HIGH MEAN EARN VOLa,t HIGH MEAN EARN VOLa,t BROKER SIZEa,t ln(EXPERIENCEa,t) ln(# INDUSTRIESa,t) MEAN ln(MKT VAL)a,t MEAN BTMa,t MEAN INST OWNa,t MEAN ROAa,t MEAN TURNa,t MEAN BUS CONCa,t MEAN RET VOLa,t MEAN ln(AGE)a,t MEAN EARN COMOVEa,t #OBS Adjusted R2 F-tests: DISTANCEi,t + DISTANCEi,t * HIGH MEAN EARN VOLi,t = 0 Values -0.298*** -0.558*** 41 Panel B. Analyst Following INTERCEPT DIST 10i,t DIST 10i,t * HIGH EARN VOLi,t [1] ln(ANAL FOLLi,t) [2] ln(ANAL FOLLi,t) 0.023 (0.196) -0.019 (-0.687) -0.088*** (-3.379) -0.012 (-0.100) # FIRMS 100i,t 0.076*** (4.746) 0.334*** (88.421) 0.167*** (14.243) 0.355*** (15.964) -0.356*** (-13.190) 7.401*** (27.518) 0.151*** (5.160) -0.331*** (-5.688) -0.106*** (-13.580) 0.054*** (3.067) 0.012 (0.440) -0.087*** (-3.348) 0.077*** (4.825) 0.335*** (88.683) 0.167*** (14.299) 0.355*** (15.927) -0.361*** (-13.371) 7.427*** (27.614) 0.154*** (5.251) -0.329*** (-5.650) -0.106*** (-13.671) 0.056*** (3.183) 33,021 0.673 33,021 0.673 # FIRMS 100i,t * HIGH EARN VOLi,t HIGH EARN VOLi,t ln(MKT VALi,t) BTMi,t INST OWNi,t ROAi,t TURNi,t BUS CONCi,t RET VOLi,t ln(AGEi,t) EARN COMOVEi,t #OBS Adjusted R2 F-tests: DISTANCEi,t + DISTANCEi,t * HIGH EARN VOLi,t = 0 Values -0.107*** -0.075*** 42 Panel C. Forecast Revision Timeliness INTERCEPT DIST 10i,t DIST 10i,t * HIGH EARN VOLi,t [1] QUICK REVi,t [2] QUICK REVi,t -0.584*** (-12.729) 0.004 (0.348) -0.045*** (-4.305) -0.586*** (-12.815) # FIRMS 100i,t 0.032*** (4.974) -0.001 (-0.681) 0.004 (0.825) 0.072*** (7.141) -0.015 (-1.313) 0.128 (1.119) 0.054*** (4.123) -0.055** (-2.065) -0.016*** (-5.116) 0.006 (0.786) 0.062*** (15.149) 0.415*** (17.608) 0.006 (0.801) 0.027*** (5.734) 0.006 (0.533) -0.053*** (-4.939) 0.036*** (5.523) -0.001 (-0.655) 0.005 (0.871) 0.072*** (7.120) -0.015 (-1.301) 0.122 (1.064) 0.054*** (4.126) -0.056** (-2.102) -0.016*** (-5.130) 0.007 (0.847) 0.062*** (15.195) 0.414*** (17.575) 0.006 (0.761) 0.028*** (5.782) 33,021 0.522 33,021 0.522 # FIRMS 100i,t * HIGH EARN VOLi,t HIGH EARN VOLi,t ln(MKT VALi,t) BTMi,t INST OWNi,t ROAi,t TURNi,t BUS CONCi,t RET VOLi,t ln(AGEi,t) EARN COMOVEi,t ln(ANAL FOLLi,t) BROKER SIZEi,t ln(# FIRMS FOLLi,t) ln(EXPERIENCEi,t) #OBS Adjusted R2 F-tests: DISTANCEi,t + DISTANCEi,t * HIGH EARN VOLi,t = 0 Values -0.041*** -0.047*** 43 Panel D. Accuracy INTERCEPT DIST 10i,t DIST 10i,t * HIGH EARN VOLi,t [1] ACCURACYi,t [2] ACCURACYi,t -0.618*** (-3.135) 0.224*** (5.722) -0.347*** (-6.283) -0.611*** (-3.129) # FIRMS 100i,t -0.111*** (-3.920) 0.048*** (4.315) -0.158*** (-4.376) 0.480*** (10.446) 0.923*** (11.465) 1.606** (2.313) -0.065 (-1.395) -4.019*** (-19.131) -0.131*** (-9.125) 0.006 (0.153) 0.224*** (11.248) 0.186 (1.398) -0.007 (-0.205) -0.045* (-1.864) 0.205*** (5.740) -0.293*** (-5.545) -0.138*** (-5.048) 0.048*** (4.286) -0.159*** (-4.402) 0.479*** (10.410) 0.917*** (11.432) 1.639** (2.359) -0.067 (-1.436) -4.022*** (-19.136) -0.131*** (-9.108) 0.008 (0.190) 0.225*** (11.301) 0.194 (1.459) -0.009 (-0.259) -0.045* (-1.875) 33,021 0.272 33,021 0.271 # FIRMS 100i,t * HIGH EARN VOLi,t HIGH EARN VOLi,t ln(MKT VALi,t) BTMi,t INST OWNi,t ROAi,t TURNi,t BUS CONCi,t RET VOLi,t ln(AGEi,t) EARN COMOVEi,t ln(ANAL FOLLi,t) BROKER SIZEi,t ln(# FIRMS FOLLi,t) ln(EXPERIENCEi,t) #OBS Adjusted R2 F-tests: DISTANCEi,t + DISTANCEi,t * HIGH EARN VOLi,t = 0 Values -0.123** -0.088* 44 TABLE 7 Industry Firm Mentions in Conference Calls Regression Results This table includes all firm/year observations from 2002 to 2011 with sufficient data to calculate the dependent and independent variables. The dependent variable in Column 1 is an indicator variable equal to 1 if firm i mentions firm j in any of its conference calls during year t (MENTIONi,j,t). In Column 2 (3), the dependent variable is an indicator variable equal to 1 if the managers (analysts) mention firm j first during any of firm i's conference calls during year t (MENTION (FIRM FIRSTi,j,t) and MENTION (ANAL FIRSTi,j,t)). The independent variable of interest is the distance between firm i and firm j in year t (DISTANCEi,j,t). All variables are defined in Appendix A. Year and industry (two-digit SIC code) fixed effects are included as additional independent variables. The coefficients on the year and industry indicator variables are suppressed. Standard errors are clustered by firm i-j combinations. All continuous variables are winsorized at the 1% and 99% levels. *, **, and *** represent significance at the 10%, 5%, and 1% levels, respectively. [1] [2] [3] MENTIONi,j,t MENTION (FIRM FIRST)i,j,t MENTION (ANAL FIRST)i,j,t INTERCEPT -12.119*** (-16.814) -12.452*** (-15.634) -13.730*** (-11.101) DISTANCEi,j,t -0.346*** (-8.233) -0.308*** (-6.511) -0.409*** (-5.831) ln(MKT VALi,t) 0.063*** (5.280) 0.032** (2.364) 0.137*** (6.324) ln(MKT VALj,t) 0.869*** (66.549) 0.904*** (61.318) 0.745*** (33.535) BTMi,t 0.022 (0.777) -0.004 (-0.132) 0.040 (0.778) BTMj,t 0.299*** (6.838) 0.314*** (6.089) 0.194** (2.552) INST OWNi,t -0.201*** (-4.223) -0.203*** (-3.686) -0.077 (-0.951) INST OWNj,t -0.912*** (-21.127) -0.954*** (-19.484) -0.712*** (-9.260) ROAi,t -0.042 (-0.771) -0.053 (-0.846) -0.038 (-0.334) ROAj,t -0.589*** (-8.260) -0.507*** (-6.095) -0.740*** (-5.331) TURNi,t 2.965*** (4.660) 3.397*** (4.784) 2.291** (2.015) TURNj,t 1.968*** (2.651) 1.445* (1.668) 4.998*** (4.125) BUS CONCi,t 0.423*** (4.548) 0.425*** (3.960) 0.489*** (3.237) 45 TABLE 7 (continued) BUS CONCj,t 0.257*** (3.402) 0.201** (2.393) 0.485*** (3.616) EARN VOLi,t 0.592*** (3.586) 0.715*** (3.940) -0.234 (-0.595) EARN VOLj,t 0.106 (0.458) 0.465* (1.780) -1.203** (-2.287) RET VOLi,t -0.104 (-0.697) -0.085 (-0.500) -0.151 (-0.472) RET VOLj,t 1.210*** (6.369) 1.229*** (5.486) 1.080*** (2.801) ln(AGEi,t) -0.042* (-1.760) -0.074*** (-2.712) 0.038 (0.931) ln(AGEj,t) 0.191*** (7.600) 0.194*** (6.892) 0.166*** (3.667) ln(ANAL FOLLi,t) 0.266*** (12.122) 0.278*** (11.156) 0.205*** (4.806) ln(ANAL FOLLj,t) 0.116*** (5.026) 0.082*** (3.198) 0.198*** (5.046) COMPETITORi,j,t 1.686*** (59.804) 1.609*** (50.546) 1.850*** (36.230) EARN COMOVEi,j,t 0.093*** (3.827) 0.061** (2.218) 0.213*** (4.516) #OBS 4,543,622 4,543,622 4,543,622 0.307 0.295 0.270 2 Adjusted R 46
© Copyright 2026 Paperzz