Firm diversification and asymmetric information: evidence from analysts’ forecasts and earnings announcements

Journal of Financial Economics 64 (2002) 373–396

Firm diversification and asymmetric information:evidence from analysts’ forecasts and

earnings announcements$

Shawn Thomas

Katz Graduate School of Business, University of Pittsburgh, Pittsburgh, PA 15260, USA

Received 26 October 1999; accepted 9 July 2001

Abstract

Managers frequently cite the desire to mitigate asymmetric information as a motivation for

increasing firm focus. The information benefits of focus appear relevant for the subset of firms

that actually increase their focus; however, the relevance of focus-related information benefits

for the population of diversified firms is an open question. This paper examines the relation

between corporate diversification and asymmetric information proxies derived from analysts’

forecasts and abnormal returns associated with earnings announcements. I find that greater

diversification is not associated with increased asymmetric information. These results call into

question the notion that corporate diversification strictly exacerbates information problems.

r 2002 Published by Elsevier Science B.V.

JEL classification: G34

Keywords: Transparency; Diversification; Focus; Asymmetric information; Corporate structure

1. Introduction

Managers frequently cite the desire to mitigate asymmetric information as amotivation for increasing firm focus.1 An implication of this motivation is that

$I thank Anwer Ahmed, Bipin Ajinkya, Sanjeev Bhojraj, Ted Fee, Mark Flannery, Charles Hadlock,

Joel Houston, Ken Lehn, M. Nimalendran, Richard Warr, an anonymous referee, and especially Mike

Ryngaert for helpful comments. I gratefully acknowledge the contributions of I/B/E/S International Inc.

and First Call for providing forecast data. All errors remain my own.

E-mail address: [email protected] (S. Thomas).1 Habib et al. (1997) and Gilson et al. (2000) cite examples in which managers appeal to information

benefits in justifying stock break-ups. Further, it seems as if focus has become a mantra of sorts among

0304-405X/02/$ - see front matter r 2002 Published by Elsevier Science B.V.

PII: S 0 3 0 4 - 4 0 5 X ( 0 2 ) 0 0 1 2 9 - 0

diversified firms are subject to larger asymmetric information problems than arefocused firms. This conjecture, which I refer to as the transparency hypothesis,suggests that splitting a conglomerate firm along industry lines into separately tradedand/or operated entities can mitigate the information asymmetries about eachindustry segment’s profitability and operating efficiency that arise because thesegments are part of a conglomerate. However, many more firms choose to remaindiversified rather than to refocus (e.g., see Montgomery, 1994; Denis et al., 1997a).

One possible explanation is that corporate diversification is not strictly associatedwith an increase in asymmetric information. The aggregate nature of reportedearnings for diversified firms could actually imply less potential for informationasymmetry. Assuming that the errors outsiders make in forecasting industry segmentcash flows are imperfectly correlated across segments, the absolute value of thepercentage error in the forecast of firm cash flows may be smaller for a diversifiedfirm than for a focused firm. In other words, even if the errors outsiders make inforecasting segment cash flows are larger than the errors they make in forecastingfocused firm cash flows, if these errors are not perfectly positively correlated, thenthe consolidated forecast may be more accurate than a forecast for a focused firm. Ineffect, asymmetric information regarding each segment’s performance is, in part,diversified away across segments. Consequently, the degree to which the expectationsof outsiders differ from managers’ private information could be reduced. I refer tothis possibility as the information diversification hypothesis.

Clearly, these hypotheses are not mutually exclusive. Thus, I empirically examinethe relation between corporate diversification and the degree of asymmetricinformation faced by outsiders. The tests use proxies for asymmetric informationderived from analysts’ forecasts and stock price reactions to earnings announce-ments. Regressions explaining analysts’ forecast errors and the dispersion amonganalysts’ forecasts reveal that greater diversification is associated with smallerforecast errors and less dispersion among forecasts, consistent with the informationdiversification effect. However, after controlling for differences in the volatility ofabnormal stock returns between diversified and focused firms (and, hence, any effectof diversification on the volatility of returns, earnings, etc.), diversification isassociated with larger forecast errors and greater dispersion among forecasts,consistent with the transparency effect.

Given the difficulty in determining the magnitude of the reduction in diversifiedfirms’ volatility that stems directly from the information diversification effect, it isdifficult to determine from the regression results which effect, if any, dominates.Thus, I compare the forecast errors of diversified firms with those of similarlyconstructed portfolios of focused firms. This analysis, by construction, allowsfocused firms to realize any benefits of the information diversification effect withoutalso incurring any of the costs associated with reduced transparency. However, I find

(footnote continued)

some executives and academics; e.g., see Bhide (1990); ‘‘Corporate Breakups Are No Panacea’’ by Roger

Lowenstein (The Wall Street Journal, June 5, 1997); or ‘‘Confessions of a Corporate-Spinoff Junkie’’ by

Roger Lowenstein (The Wall Street Journal, March 28, 1996).

S. Thomas / Journal of Financial Economics 64 (2002) 373–396374

that diversified firms’ forecast errors are very similar in magnitude to those of theirmatching-firm portfolios.

Finally, diversified firms are shown to have smaller revaluations associated withearnings announcements, consistent with outsiders being better able to anticipate theearnings of diversified firms. Additionally, diversified firms have slightly largerearnings response coefficients (ERCs), suggesting that new earnings information fordiversified firms is capitalized into stock prices to a greater extent than new earningsinformation for focused firms. Taken together, the results in this paper call intoquestion the notion that corporate diversification strictly exacerbates informationproblems for most firms. Further, these results suggest that the firms most likely torealize information benefits from undertaking stock break-ups are those withparticularly severe diversification-related transparency problems.2

The rest of this paper is organized as follows. Section 2 develops the hypotheses tobe tested. Section 3 describes the sample and empirical design of the tests. Section 4presents the empirical results. Section 5 offers a concluding discussion.

2. Hypotheses development and literature review

The degree of information asymmetry between managers and outsiders may differfor diversified versus focused firms. The source of the difference in asymmetry couldbe that diversified firms are less transparent than focused firms. For instance, whilemanagers of diversified firms can observe divisional cash flows, outsiders can observeonly noisy estimates of divisional cash flows. Thus, the mapping of divisional cashflows into consolidated earnings can be less than transparent to outsiders, andreported earnings will convey less value-relevant information. To the extent thataccounting figures for diversified firms are less transparent relative to those offocused firms, it is possible that asymmetric information problems are more severefor diversified firms.

In addition, diversified firms often operate in many different industries whileindividual financial analysts often specialize within one particular industry.Following a diversified firm will by definition take an analyst out of his or herarea of expertise at least along some dimension. Thus, we might expect individualanalysts to have greater difficulty in following diversified firms. Dunn and Nathan(1998) find that as the number of diversified firms a particular analyst followsincreases, the accuracy of that analyst relative to other analysts declines. To theextent that the complexity of diversified firms might reduce the effectiveness ofindividual analysts in processing information about firms, larger informationasymmetries between managers and investors could exist.

I refer to the idea that aggregated cash flows and other diversification-relatedinformation problems make it more difficult for analysts (outsiders) to forecast firmcash flows as the transparency hypothesis. The transparency hypothesis predicts

2 Diversification-related transparency problems could represent the inability of segments of diversified

firms to attract industry specialist analysts as in Gilson et al. (2000).

S. Thomas / Journal of Financial Economics 64 (2002) 373–396 375

that, compared with focused firms, diversified firms should have, all else equal, largerforecast errors, more dispersion among analysts’ forecasts, larger revaluationsaround earnings announcements, and smaller ERCs. To the extent that they are lesstransparent than focused firms, diversified firms will face more difficulty in raisingcapital, less stock market liquidity, and, therefore, higher costs of capital. Managersof diversified firms could reduce the information gap by credibly increasing segmentdisclosure or by breaking up firms along industry lines into separately tradedsecurities and/or separately operated firms.3

The benefits of increased disclosure have been examined extensively in previousliterature (e.g., see Diamond and Verrecchia, 1991). Lang and Lundholm (1996) findthat firms with more informative disclosure policies as measured by ratings from theFinancial Analysts Foundation have larger analyst followings, less dispersion amongindividual analysts’ forecasts, and less volatility in forecast revisions. Swaminathan(1991) finds that multiple-segment firms experienced increases in the accuracy ofanalysts’ forecasts and decreases in the dispersion among analysts’ forecastssubsequent to implementation of the SEC’s line of business disclosure requirements.Piotroski (1999) finds that discretionary expansion of segment reporting is also, onbalance, associated with an increase in analysts’ forecast accuracy and a decrease inforecast dispersion. These findings are all consistent with increased disclosurereducing information asymmetries.

In addition to improving disclosure, managers can break up the conglomerate firmalong industry lines into separately traded and/or operated entities to improvetransparency. For instance, a frequently cited motivation for focus-increasingtransactions and tracking stock issues is management’s desire to make theircompanies easier for investors to evaluate. A direct implication of this rationale isthat the potential for information asymmetry between managers and outsiders isgreater for more-diversified firms. Habib et al. (1997) present a model in whichsplitting a firm along industry lines into separately traded firms leads to moreinformative stock prices, in turn improving the quality of managers’ investmentdecisions and reducing uninformed investors’ uncertainty about asset values.Similarly, Nanda and Narayanan (1999) present a model of optimal corporatescope in which managers trade off the benefits of internal capital markets againstdiversification-related asymmetric information costs.

Gilson et al. (2000) document increases in analysts’ forecast accuracy andconsensus for those firms that conduct stock break-ups, i.e., spinoffs, carve-outs, ortargeted stock offerings. These improvements appear to stem from the enhancedability of the newly created focused entities to attract industry-specialist analysts.Zuta (1998) reports improved correspondence between the sum of segment cashflows and consolidated firm cash flows for diversified firms subsequent to targetedstock issues. Krishnaswami and Subramaniam (1999) find that, relative to controlfirms, diversified firms that conduct spinoffs have larger forecast errors, greaterdispersion among forecasts, and larger revaluations around earnings announce-

3 Targeted stock allows parent companies to retain legal ownership of business segments while holders of

targeted stock are entitled to the earnings stream of a particular business segment, e.g., see Zuta (1998).


ments. Krishnaswami and Subramaniam also find that firms that engage in spin-offssubsequently raise more external capital than control firms. This finding is consistentwith some firms conducting spinoffs in an effort to reduce asymmetric informationprior to approaching capital markets.

While the stock break-up results provide evidence in support of the transparencyhypothesis, it is important to recognize the potential limitations of these papers.First, all of these studies, by definition, consider only firms that changeorganizational form. These are precisely the subsets of firms that would be expectedto have suffered the greatest diversification-related information problems. Second, itis not clear whether the source of the reduction in information asymmetry isimproved transparency. Denis et al. (1997b) report that decreases in diversificationare frequently associated with external corporate control threats, financial distress,and management turnover. Thus, the improvements in asymmetric informationproxies around break-ups could be due to the resolution of uncertainty regardingthese events’ potential impact on earnings. By examining changes at the margin,these papers cannot address the fundamental question of how firm diversificationand asymmetric information are related for the broad cross-section of firms that donot choose (or are not forced) to refocus or issue tracking stock (or diversify, for thatmatter). These papers suffer from a sample selection bias in this regard. Hence, thefindings might not generalize beyond the respective samples.

Unlike previous research, this paper also explicitly recognizes the potentialbenefits of diversification in the context of analysts’ forecast characteristics andreactions to earnings surprises. The aggregate nature of reported earnings fordiversified firms could imply less potential for information asymmetry. For example,assuming that the errors outsiders make in forecasting industry segment cash flowsare imperfectly correlated across segments, the absolute value of the percentage errorin the forecast of firm cash flows may be smaller for a diversified firm than for afocused firm. In other words, even if the errors outsiders make in forecasting segmentcash flows are larger than the errors they make in forecasting focused firm cashflows, if these errors are not perfectly positively correlated, then the consolidatedforecast may be more accurate than a forecast for a focused firm. Consequently, thedegree to which the expectations of outsiders differ from managers’ privateinformation could be reduced. Again, I term this possibility the informationdiversification hypothesis.

Subrahmanyam (1991) and Gorton and Pennacchi (1993) draw on similarreasoning to explain the information benefits of trading baskets of stocks rather thanthe individual stocks that constitute the baskets. They demonstrate that basketsecurities are subject to less asymmetric information (and, hence, adverse selection)precisely because their cash flows are aggregates. In effect, information asymmetriesassociated with each security are, in part, diversified away across the securities thatconstitute the basket. Thus, information asymmetries are often lower for the basketsecurity than for the individual securities constituting the basket. To the extent thatdiversified firms represent basket securities consisting of focused firms, these samearguments imply that diversified firms may be less subject to asymmetric informationproblems than focused firms.


This reasoning is consistent with anecdotal evidence in the financial press. Forexample, General Electric (GE) operates in many different industries, and asignificant fraction of GE’s assets are opaque financial assets. These factors shouldcombine to make GE less transparent. However, Ben Zacks of Zacks InvestmentResearch indicates that analysts’ forecasts for GE actually fall in a ‘‘very, very tightrange’’. While GE is known for providing ‘‘guidance’’ to analysts, individualanalysts also appeal to the fact that GE is a well-diversified portfolio to explain theremarkable accuracy of and lack of dispersion among their forecasts for GE.Similarly, the recent Time Warner and America Online merger illustrates theintuition behind the information diversification hypothesis.4 While analystsfollowing the combined company are now charged with evaluating quite disparatebusinesses, the possibility that forecast errors across these businesses will ‘‘balanceout’’ is advanced as a mitigating factor.

Hadlock et al. (2001) examine the relation between firm diversification andasymmetric information in the context of seasoned equity offerings. They find thatequity issue announcements by diversified firms are met with significantly less-negative revaluations than announcements by focused firms. Haw et al. (1994) findthat analysts’ forecast accuracy decreases for firms immediately subsequent tocompleting an acquisition but returns to pre-acquisition levels within several years.However, the temporary reduction in accuracy is significantly less pronounced forthe subset of diversifying acquisitions due to the reduction in post-merger earningsvolatility that results from combining cash flow streams that have lower correlations.Fee and Thomas (2001) find a robust negative relation between a firm’s level ofunrelated diversification and measures of asymmetric information based on stockmarket trading characteristics. Taken together, these results are consistent withdiversification alleviating information problems.

In sum, the information diversification hypothesis predicts that diversified firmsshould have smaller forecast errors, less dispersion among analysts’ forecasts,smaller market reactions to earnings announcements, and larger ERCs than focusedfirms.

3. Data and empirical design

3.1. Description of sample

I draw my initial sample of analyst forecast and actual earnings data from I/B/E/S.As the consensus forecast, I use the median forecast for earnings per share (EPS)reported for the month closest to, but preceding, the month in which actual earningsare released. I require that each sample firm have a minimum of three analysts

4 See ‘‘Managing Profits: How General Electric Damps Fluctuations In Its Annual Earnings’’ by

Randall Smith et al. (The Wall Street Journal, November 3, 1994), and ‘‘AOL and Time Warner Leaves

Street Guessing on New Animal’s Value’’ by Paul Sherer and Elizabeth MacDonald (The Wall Street

Journal, January 13, 2000).


providing forecasts. The forecast data are for fiscal years ending between July 1985and June 1996. For all firms with the necessary I/B/E/S data, I identify those firmsthat also have data available from Compustat on both a consolidated and industrysegment basis (both research and active files). All variables constructed fromCompustat data are measured at the end of the fiscal year that precedes the year ofthe forecast. I exclude firms with reported business segments that are regulatedutilities (SIC codes in the range 4800–4829 and 4910–4949) or financial servicesoperations (SIC codes between 6000 and 6999). For the remaining firms, I identifythose with data available from the Center for Research in Security Prices (CRSP)database. I then exclude foreign firms, American Depository Receipts (ADRs), andreal estate investment trusts (REITs). From this sample, I then exclude thoseobservations for which the absolute forecast error, defined as jactual EPS�medianforecast EPSj; exceeds 100% of the stock price five days before earnings areannounced (e.g., see Ali et al., 1992). I also exclude those observations for which thestandard deviation of analysts’ forecasts is greater than 20% of the stock price.5

In order to obtain consistency between the stock prices (from CRSP) used asdeflators and I/B/E/S EPS data, I restate actual and forecasted EPS using the I/B/E/S stock split adjustment factors to recover pre-split forecasts and actuals. Thus, theEPS data in my sample reflect the numbers reported when the forecasts were made,i.e., not adjusted for subsequent splits. A total of 12,282 forecast years remain afterall screens.

Table 1 provides details on the characteristics of the sample. The frequency ofobservations by forecast year and firm type is reported in Panel A. While the numberof observations per year grows over the sample period, much of this growth stemsfrom Compustat and I/B/E/S expanding the coverage of their databases. Forexample, I/B/E/S first began covering Nasdaq firms in 1984 and has since expandedits coverage of Nasdaq firms. Thus, many firms arriving on the databases over thisperiod have high levels of focus as evidenced by the increase in single-segment firmobservations.6

The frequency of observations by firm, industry, and fiscal year is reported inPanel B. There are 2,677 distinct firms represented, each of which is included anaverage (median) of seven (eight) times. These firms operate in 224 distinct industries(based on three-digit SIC codes) with a mean (median) individual industryrepresentation of 192 (88) firm-years per industry. There are 11 fiscal years includedin the sample, with a mean (median) of 1,189 (1,133) observations per year.

5 The forecast error screen results in the deletion of 70 observations (55 single segment and 15 multiple

segment). The frequency of extreme observations is significantly different (10% level) across firm type. The

dispersion screen results in the deletion of only 29 observations (19 single segment and 10 multiple

segment). To ensure the results are not driven by this criteria, I tried several other cutoffs (e.g., standard

deviation greater than 10%, 25%, or 30% of stock price). The results are not sensitive to the criteria used.6 Comment and Jarrell (1995) provide evidence regarding the focus levels of firms arriving on Compustat

during the 1980s.


3.2. Diversification measure

I use an asset-based Herfindahl Index (HERF) as a measure of diversification.This variable reflects the degree to which the assets of a firm are concentrated withinits industry segments. For each firm, I calculate HERF as the sum of the squares ofeach reported segment’s assets as a proportion of the firm’s total assets. Thus, forfirm i in year t; HERF is measured as

HERFit ¼XNit

j¼1

TAjit

XNit

j¼1

TAjit

, !2

; ð1Þ

where Nit is the number of reported segments of firm i at time t and TAjit are theassets attributable to segment j of firm i at time t: HERF equals one for all single-segment firms and is less than one for multiple-segment firms, where smaller levels of

Table 1

Sample characteristics

This sample includes firms with analyst forecast and earnings data available from I/B/E/S for fiscal years

ending between July 1985 and June 1996. Firms must be followed by at least three analysts and also have

data available from Compustat and CRSP. Firms with reported business segments that are regulated

utilities (SIC codes in the range 4800–4829 and 4910–4949) or financial services operations (SIC codes

between 6000 and 6999) are excluded, as are foreign firms, ADRs, and REITs. Additionally, observations

for which the absolute forecast error, jactual�median forecastj; is greater than 100% of the stock price or

for which the standard deviation of analysts’ forecasts is greater than 20% of the stock price are also

excluded. Industries in Panel B are defined using three-digit SIC codes.

Panel A: Frequency of observations by forecast year and firm type

Forecast year Multiple segment Single segment Total

1985 326 366 692

1986 341 419 760

1987 354 518 872

1988 328 609 937

1989 342 696 1,038

1990 365 766 1,131

1991 342 791 1,133

1992 337 868 1,205

1993 354 1,021 1,375

1994 364 1,172 1,536

1995 361 1,242 1,603

Totals 3,814 8,468 12,282

Panel B: Frequency of observations as a function of firm, industry, and year

Number of times represented

Total Minimum Maximum Mean Median Standard deviation

Firm 2,677 1 11 7 8 3.40

Industry 224 1 656 192 88 200.43

Year 11 692 1,603 1,189 1,133 284.86


HERF correspond to less concentration of assets among segments and hence greaterdiversification.7

3.3. Forecast accuracy and dispersion

I use the accuracy of consensus forecasts and the dispersion among forecasts asproxies for asymmetric information.8 These proxies are based on forecasts made inthe month before actual earnings are released. I choose to analyze forecastcharacteristics for the shortest possible forecasting horizon in order to minimize theoptimism bias that appears to exist in forecasts made at the beginning of a fiscal year(e.g., see O’Brien, 1988; Easterwood and Nutt, 1999). Additionally, Elton et al.(1984) demonstrate that the errors in forecasts made very near to the end of aforecasting period consist primarily of firm-specific information rather thaneconomy- or industry-wide information. These factors make forecasts very nearthe end of a forecasting period attractive as proxies for assessing differences inasymmetric information across firms.

As the primary measure of forecast accuracy, I use ERROR, which is the absolutedifference between actual earnings and the median forecast deflated by the stockprice five days before the earnings announcement date. Firms with larger differencesin information asymmetry between managers and outsiders regarding firm earningsare expected to have larger forecast errors. DISPERSION is the standard deviationof analysts’ forecasts deflated by the stock price five days before the earningsannouncement date. This variable is a measure of disagreement among analysts.Disagreement could result from a lack of available information about a firm. Thus,greater disagreement among analysts’ forecasts could imply larger informationproblems.

In order to examine the relation between diversification and informationasymmetry, I regress forecast accuracy and dispersion on HERF. The transparencyhypothesis predicts that higher HERF, or greater focus, should be associated with

7 As an alternative to HERF, I use the Jacquemin and Berry (1979) entropy measure of total

diversification, which can be decomposed into separate measures of related and unrelated diversification.

The results using the unrelated diversification measure are very similar to those reported; however,

consistent with the intuition underlying the information diversification hypothesis, I find no significant

relation between the measures of asymmetric information and the degree of related diversification.

Additionally, I obtain results similar to those reported when I use Herfindahls calculated at the three- or

two-digit SIC code level, a multiple-segment dummy, or the disclosed number of segments as

diversification proxies.8 Other authors have used analyst following as a proxy for the supply of information about a firm (e.g.,

see Brennan and Subrahmanyam, 1995). However, in equilibrium the number of analysts following a firm

can be either positively or negatively related to the level of asymmetric information surrounding the firm.

While I have examined the relation between firm diversification and analyst following for the firms in my

sample, I do not report these results given the difficulty of making inferences about differences in

asymmetric information from analyst followings. My results are similar to those reported in Bhushan

(1989), i.e., all else equal, diversified firms are followed by fewer analysts. If one interprets smaller analyst

followings as an indication of increased information asymmetry, then this relation should bias against

finding support for the information diversification hypothesis.


smaller forecast errors and less dispersion among forecasts. Alternatively,the information diversification hypothesis predicts that higher HERFshould be associated with larger forecast errors and greater dispersion amongforecasts.

In order to draw appropriate inferences from the forecast analysis, I also controlfor other factors that can impact forecast characteristics. For example, firm size canbe expected to increase forecast accuracy and reduce forecast dispersion (e.g., seeAtiase, 1985). Thus, I include the book value of total assets (TA) at the end of theprevious fiscal year.

As in Alford and Berger (1999), I include a measure of stock returnvolatility (VOLATILITY), calculated as the standard deviation of the marketmodel residuals over the period from 210 to 11 days before the earningsannouncement date. The volatility of stock prices proxies for the amount of price-relevant information about a firm that arrives daily to the market. Alford and Bergerargue that as volatility increases, the amount of price-relevant informationthat analysts must process also goes up and analysts’ ability to forecastearnings declines. Thus, firms with higher volatility are expected to have largerforecast errors and more analyst disagreement. Given that diversified firms generallyexhibit less volatility than focused firms (e.g., see Comment and Jarrell, 1995),volatility differences between diversified and focused firms could in part reflect theinformation diversification effect described in Section 2. Thus, it will be important toaccount for the relation between volatility and diversification in interpreting theresults below.

Analysts might face more difficulty in forecasting earnings for firms with a lot ofpotential growth options relative to firms whose values consist mainly of assets-in-place. Thus, I also include in the analysis the ratio of R&D expense to sales(RDSALES) and the ratio of intangible assets to total assets at the previous fiscalyear end (INTGTA).9 Barth et al. (1998) conjecture that the level of analyst effortand, perhaps by extension, the quality of analysts’ forecasts vary with the degree towhich firm value is comprised of tangible assets.

Since leverage adds to the volatility of earnings, the ratio of long-term debt anddebt in current liabilities to total assets (LEVG) is included.10 Firms with higherleverage might be expected to have less-accurate forecasts and more dispersionamong forecasts. I also include a binary variable (SALESDEV) that takes the valueof one if the sum of segment sales is not within 1% of consolidated firm sales. Thisvariable is intended to gauge the difficulty outsiders face in aggregating segment cashflows for diversified firms. More difficulty in mapping segment cash flows intoconsolidated cash flows may imply more analyst disagreement and less forecastaccuracy.

9 For firms without R&D expense on Compustat, I set R&D to zero and a binary variable denoting that

this data item is missing is set to one. The missing R&D dummy is never large in magnitude or statistically

significant and is not included in the results below. Intangible assets also include goodwill.10 In the 39 observations where leverage is greater than one, LEVG is set equal to one.


3.4. Price impact of earnings surprises

Dierkens (1991) argues that, all else equal, a strong market reaction to an earningsannouncement of a given firm is an indication that the managers of that firm havereleased substantial private information and information asymmetry betweeninsiders and outsiders is large for that firm. Similarly, in models of time-varyingasymmetric information, the periods immediately after earnings announcements areoften characterized as being periods of low information asymmetry, consistent withthe announcement eliminating much of the asymmetry that existed prior to therelease. For example, Korajczyk et al. (1991) find that the stock price declines atannouncements of seasoned equity offerings are increasing in the time since the lastearnings release.

To assess the informativeness of earnings announcements for stock prices, I useevent study methodology to estimate abnormal returns (ARs) for three-day windowscentered on the annual earnings announcement dates from I/B/E/S. In estimating themarket model, I use the firm’s daily returns and the returns on the CRSP value-weighted index over days �210 to �11 where day 0 is the earnings announcementdate. Since I am primarily interested in the magnitude of the stock price response andnot the direction, I use the absolute abnormal return, jARj; in the tests below.

I regress jARj on HERF and other variables that could impact the market’sreaction to an earnings announcement. The transparency hypothesis predicts largerreactions to earnings surprises for diversified firms relative to focused firms, or anegative relation between jARj and HERF. This would be consistent with outsidersbeing better able to anticipate the earnings of focused firms. Alternatively, theinformation diversification hypothesis predicts a positive relation between jARj andHERF. Clearly, the magnitude of the earnings surprise and the ex ante disagreementamong analysts will influence the market’s reaction. Thus, the forecast error anddispersion measures are included in the specifications. Larger differences between themarket’s expectations of earnings and actual earnings are expected to elicit biggerreactions from the market, i.e., larger ERCs as in Beaver et al. (1979). Greaterdisagreement among analysts is also expected to result in larger market reactions toearnings announcements.

I also include an interaction term of HERF and ERROR to assess the impact ofdiversification on the ERC. The transparency hypothesis would predict that focusedfirms should have earnings surprises that are capitalized to a greater extent into stockprices than the earnings surprises of diversified firms and, hence, a positive coefficienton the interaction term. Alternatively, the information diversification hypothesiswould predict that the coefficient on the interaction term would be negative,consistent with earnings surprises of diversified firms being more value relevant.

The other variables used in the analysis to explain abnormal returns are drawnprimarily from the ERC literature. Easton and Zmijewski (1989), among others,demonstrate that constraining the coefficient on ERROR (i.e., the ERC) to be thesame across all firms can lead to other explanatory variables having explanatorypower only because they are correlated with the cross-sectional variation in the ERC.Thus, I include interaction terms of the ERC and controls for cross-sectional


determinants of the ERC identified by the literature. Primary among thesedeterminants are growth opportunities, firm size, and risk. As a proxy for thequality of investment opportunities, I include the market-to-book ratio (MB),defined as the ratio of the firm’s market value (market value of equity plus the bookvalue of total assets minus the book value of equity) to the firm’s book value of totalassets at the prior fiscal year-end. As a measure of firm size, I include market value ofequity at the end of the previous fiscal year (MVE). Finally, I include the ratio ofdebt to total assets (LEVG) as a proxy for risk.

4. Results

4.1. Univariate comparisons

Table 2 reports descriptive statistics grouped by whether the firm has multiplesegments or a single segment. While the dispersion among analysts’ forecasts issimilar for single- and multiple-segment firms, the accuracy of the forecasts is quitedifferent across firm type. The mean absolute forecast error is 3.71% of the stockprice for multiple-segment firms and 4.59% for single-segment firms. Thus, analysts’earnings forecasts for multiple-segment firms are significantly more accurate thanthose for single-segment firms. Further, the market’s reaction to an earnings report issignificantly larger for single-segment firms than for multiple-segment firms. Themean absolute abnormal return to an earnings announcement is 3.47% for amultiple-segment firm and 4.96% for a single-segment firm. Taken together, theseresults suggest larger information asymmetries for focused firms. However, in thetests below, I attempt to control for other potential differences between diversifiedand focused firms that could also affect information asymmetry but are ignored inthese univariate comparisons.

While the multiple-segment firms in the sample exhibit varying levels of industryfocus, the median HERF is 0.51. This roughly corresponds to a two-segment firmwith 60% of its assets in one industry and 40% of its assets in another industry.Consistent with portfolio theory, multiple-segment firms have less volatility inabnormal returns. Also, multiple-segment firms are generally larger than single-segment firms. Multiple-segment firms engage in less R&D as a percentage of sales,and, as in Lang and Stulz (1994), multiple-segment firms have lower market-to-bookratios than single-segment firms. Multiple-segment firms also utilize more leveragethan single-segment firms and report more intangible assets as a fraction of totalassets. Finally, segment sales do not sum to within 1% of consolidated sales fornearly 11% of multiple-segment firm-years.

4.2. Forecast accuracy and dispersion regression results

The regression results using HERF to explain analysts’ forecast errors arereported in Table 3. In the first column of Table 3, the only explanatory variablebesides year dummies is HERF. The coefficient on HERF is positive and highly


significant. Thus, it appears that analysts’ forecasts for focused firms are lessaccurate than for diversified firms. This is consistent with the informationdiversification hypothesis and inconsistent with the transparency hypothesis. Recallthat these hypotheses are not mutually exclusive. Thus, a cautious interpretation ofthe positive coefficient on HERF is that the information diversification effectdominates the transparency effect in terms of the net effect on analysts’ forecastaccuracy.

Given the correlation between diversification and other firm characteristicsdocumented in Table 2, I introduce additional control variables that might influenceanalysts’ forecast accuracy. In Column 2 of Table 3, ln(TA) and RDSALES areadded. The negative coefficient on firm size is consistent with larger firms having

Table 2

Descriptive statistics by firm type

All I/B/E/S forecast data are for the month closest to, but preceding, the annual earnings announcement.

ERROR is the absolute forecast error, jactual�median forecastj; scaled by the firm’s stock price five days

before the earnings announcement. DISPERSION is the standard deviation of analysts’ forecasts scaled

by the firm’s stock price five days before the earnings announcement. jARj is the absolute value of the

abnormal return for a three-day window centered on the earnings announcement date calculated from a

market model estimated over the period from 210 to 11 days before the earnings announcement date.

HERF is the asset-based Herfindahl Index. VOLATILITY is the standard deviation of market model

residuals over the period from 210 to 11 days before the earnings announcement date. TA is the total assets

reported at the previous fiscal year-end in millions of dollars. MVE is the market value of a firm’s equity at

the end of the previous fiscal year in millions of dollars. RDSALES is the ratio of R&D expense to sales at

the previous fiscal year-end. MB is the ratio of the firm’s market value (market value of equity plus the

book value of total assets minus the book value of equity) to the firm’s book value of total assets at the

prior fiscal year-end. LEVG is the ratio of long-term debt and debt in current liabilities to total assets.

INTGTA is the ratio of intangible assets to total assets at the previous fiscal year-end. SALESDEV is a

binary variable that takes the value of one if the sum of segment sales is not within 1% of consolidated

firm sales for the previous fiscal year. Differences in medians are assessed using a Wilcoxon rank-sum test.

Multiple segment Single segment

ðn ¼ 3; 814Þ ðn ¼ 8; 468Þ Differences

Mean Median Mean Median Means Medians

ERROR 3.71% 1.21% 4.59% 1.24% � 0:88nnn � 0:03%nnn

DISPERSION 1.02% 0.53% 1.03% 0.49% � 0.01% 0.04%

jARj 3.47% 2.43% 4.96% 3.40% � 1:49%nnn � 0:97%nnn

HERF 0.5252 0.5112 1.0000 1.0000 � 0.4748 � 0.4888

VOLATILITY 1.97% 1.74% 2.83% 2.61% � 0:86%nnn � 0:87nnn

TA 2,834.65 873.12 733.13 184.01 2,101.52nnn 689:11nnn

MVE 2,410.57 680.11 835.31 219.02 1,575.26nnn 461:09nnn

RDSALES 0.0186 0.0005 0.0775 0.0000 � 0:0589nnn 0:0005nnn

MB 1.57 1.35 2.14 1.67 � 0:57nnn � 0:32nnn

LEVG 0.2487 0.2330 0.1922 0.1540 0:0565nnn 0:0790nnn

INTGTA 0.0627 0.0158 0.0438 0.0000 0:0189nnn 0:0158nnn

SALESDEV 0.1067 0.0000 0.0000 0.0000 0.1067 0.0000

nnn; nn; and n denote significance at the 0.01, 0.05, and 0.10 level, respectively.


smaller forecast errors. The level of R&D spending does not appear related to theaccuracy of analyst’s forecasts in this specification.

In Column 3 of Table 3, LEVG is included. The coefficient on this variable ispositive and significant indicating that firms with higher leverage have less accurateanalysts’ forecasts. In Column 4 of Table 3, the natural log of one plus the ratio ofintangible assets to total assets and SALESDEV are included. Firms with moreintangible assets appear to have more accurate analysts’ forecasts as evidenced by thenegative and significant coefficient on ln(1+INTGTA). Finally, firms whosesegment cash flows do not sum to consolidated cash flows do not appear to haveless accurate forecasts.

In Column 5 of Table 3, the volatility measure is added. Volatility is an importantdeterminant of forecast accuracy as evidenced by the dramatic increase in adjustedR2: As expected, firms with greater volatility have larger forecast errors.Interestingly, the coefficient on HERF is now negative and significant. Note that

Table 3

Panel regressions of analysts’ forecast errors on firm characteristics

The dependent variable is ERROR, the absolute forecast error, jactual�median forecastj; scaled by the

firm’s stock price five days before the earnings announcement. HERF is the asset-based Herfindahl Index.

Ln(TA) is the natural log of total assets reported at the previous fiscal year-end. RDSALES is the ratio of

R&D expense to sales at the previous fiscal year-end. LEVG is the ratio of long-term debt and debt in

current liabilities to total assets. Ln(1+INTGTA) is the natural log of one plus the ratio of intangible

assets to total assets at the previous fiscal year-end. SALESDEV is a binary variable that takes the value of

one if the sum of segment sales is not within 1% of consolidated firm sales for the previous fiscal year.

VOLATILITY is the standard deviation of market model residuals over the period from 210 to 11 days

before the earnings announcement date. The regressions also include dummy variables for years 1986–

1995. The estimated coefficients for the year dummies are not reported below. t-Statistics based on the

White-adjusted standard errors are in parentheses ðN ¼ 12; 282Þ:

Regression (1) (2) (3) (4) (5)

INTERCEPT 3:12nnn 7:56nnn 7:56nnn 7:62nnn � 8:86nnn

(7.52) (12.27) (12.46) (12.46) ð�11:08ÞHERF 2:55nnn 0:91nnn 0:96nnn 0:86nn � 1:01nnn

(8.65) (2.73) (2.88) (2.49) ð�3:14Þln(TA) � 0:56nnn � 0:82nnn � 0:82nnn 0:94nnn

ð�10:49Þ ð�14:29Þ ð�14:26Þ (11.96)

RDSALES 0.09 0:67nn 0:58n � 4:01nnn

(0.30) (2.13) (1.85) ð�10:31ÞLEVG 9:10nnn 9:51nnn 5:18nnn

(14.93) (15.36) (9.68)

ln(1+INTGTA) � 4:84nnn � 2:80nnn

ð�5:23Þ ð�3:53ÞSALESDEV 0.28 � 0.43

(0.89) ð�1:47ÞVOLATILITY 4:37nnn

(22.82)

Adjusted R2 0.01 0.02 0.05 0.05 0.25



Comment and Jarrell (1995) show that diversification helps to significantly reducevolatility. Hence, it is likely that a reduction in asymmetric information consistentwith the information diversification effect described in Section 2 would partlymanifest itself in the form of reduced stock return volatility. It should also be notedthat the coefficients on firm size and R&D to sales also change sign when volatility isincluded in the specification.

The results in Table 3 appear consistent with the following interpretation.Diversification increases the accuracy of analysts’ forecasts in part through its effecton volatility, but, after completely controlling for this effect, diversification decreasesthe accuracy of analysts’ forecasts due to the transparency problems that accompanydiversification.11 Unfortunately, given that it is unclear how much of the reduction involatility is solely due to diversification, it is difficult to conclude from Table 3 whicheffect, if any, dominates. However, it does appear that absent any significantreduction in volatility due to diversification, diversification is associated with largerforecast errors perhaps because of reduced transparency. Further examination ofthese effects is the focus of Section 4.3.

The regression results using HERF to explain dispersion among analysts’ forecastsare reported in Table 4. In the first column of Table 4, the only explanatory variablebesides year dummies is HERF. The coefficient on HERF is positive and highlysignificant. Since increases in HERF represent increases in focus, it appears thatanalysts’ forecasts are more dispersed for focused firms than for diversified firms.

In Columns 2–4 of Table 4, control variables with the exception of volatility areadded. The controls affect dispersion among forecasts much as they affect forecastaccuracy. The negative coefficient on firm size is consistent with larger firms havingless forecast dispersion. Firms with higher R&D expenditures also have greaterforecast dispersion. Firms with higher leverage have more dispersion amonganalysts’ forecasts, while firms with more intangible assets appear to have lessdispersion among analysts’ forecasts. Finally, the mapping of segment cash flowsinto consolidated cash flows does not appear to influence the dispersion amonganalysts’ forecasts. The coefficient on HERF remains positive and significant in eachof these specifications.

In Column 5 of Table 4, stock return volatility is added. As with the forecastaccuracy results above, volatility is an important determinant of forecast dispersionas evidenced by the dramatic increase in adjusted R2: Firms with greater volatilityhave more dispersion among analysts’ forecasts. The coefficient on HERF isnegative and significant after completely controlling for volatility differencesbetween diversified and focused firms. Again, absent any reduction in volatilityaccompanying diversification, diversification is associated with greater dispersion

11 Dunn and Nathan (1998) report similar results. However, the authors interpret the coefficient on the

diversification variable as evidence that diversified firms are subject to larger information asymmetries.

This interpretation does not consider the documented benefit of diversification in terms of reduced

volatility. Dunn and Nathan conjecture that the benefits of reduced volatility are only likely to be

substantial in special cases where the correlation between divisional cash flows is negative. However,

portfolio theory suggests these benefits are present anytime divisional cash flows are not perfectly

positively correlated.


among analysts’ forecasts. Note that the coefficient on firm size again changes signafter including volatility.

The results reported in Tables 3 and 4 are robust to logical cuts of the sample (e.g.,diversified firms, firms that survive at least five years, firms above/below median size,etc.). The results are also robust to excluding those firm-years in which First Callindicates that management issued earnings guidance.12 Finally, to mitigate potentialproblems with errors being cross-sectionally correlated, I estimate the regressionequations using the Fama and MacBeth (1973) methodology. The magnitudes, signs,

Table 4

Panel regressions of dispersion among analysts’ forecasts on firm characteristics

The dependent variable is DISPERSION, the standard deviation of analysts’ forecasts scaled by the firm’s

stock price five days before the earnings announcement. HERF is the asset-based Herfindahl Index.

Ln(TA) is the natural log of total assets reported at the previous fiscal year-end. RDSALES is the ratio of

R&D expense to sales at the previous fiscal year-end. LEVG is the ratio of long-term debt and debt in

current liabilities to total assets. Ln(1+INTGTA) is the natural log of one plus the ratio of intangible

assets to total assets at the previous fiscal year-end. SALESDEV is a binary variable that takes the value of

one if the sum of segment sales is not within 1% of consolidated firm sales for the previous fiscal year.

VOLATILITY is the standard deviation of market model residuals over the period from 210 to 11 days

before the earnings announcement date. The regressions also include dummy variables for years 1986–

1995. The estimated coefficients for the year dummies are not reported below. t-Statistics based on the

White-adjusted standard errors are in parentheses ðN ¼ 12; 282Þ:

Regression (1) (2) (3) (4) (5)

INTERCEPT 0:87nnn 0:79nnn 0:79nnn 0:82nnn � 1:74nnn

(11.07) (6.47) (6.61) (6.87) ð�11:46ÞHERF 0:29nnn 0:17nnn 0:18nnn 0:13nn � 0:16nnn

(5.49) (2.66) (2.84) (2.01) ð�2:82Þln(TA) 0.01 � 0:05nnn � 0:04nnn 0:23nnn

(0.68) ð�4:74Þ ð�4:61Þ (15.65)

RDSALES 1:53nnn 1:65nn 1:62nnn 0:91nnn

(10.27) (11.04) (10.81) (6.69)

LEVG 1:97nnn 2:13nnn 1:45nnn

(16.61) (17.51) (13.53)

ln(1+INTGTA) � 1:92nnn � 1:61nnn

ð�11:90Þ ð�11:04ÞSALESDEV 0.04 � 0.07

(0.59) ð�0:95ÞVOLATILITY 0:68nnn

(25.54)

Adjusted R2 0.01 0.05 0.08 0.09 0.23


12 Management guidance as documented by First Call is not very prevalent over the sample period.

Among firms in my sample, 1% of diversified firm-years and 2% of focused firm-years have a management

forecast entered in First Call. The difference in frequency of management guidance across the two

subsamples is significant at the 1% level ðt ¼ �2:92Þ: The increased frequency of management forecasts

among focused firms should bias against finding support for the information diversification hypothesis.


and significance of all coefficients are very similar to those reported in the pooledtime-series, cross-sectional results of Tables 3 and 4.

4.3. Forecast errors for similarly constructed portfolios of focused firms

The tests in Section 4.2 have several potential shortcomings. First, while the resultsare consistent with the intuitive conclusion that diversification alleviates informationproblems via a reduction in volatility while exacerbating information problems dueto transparency, it is difficult to determine which effect, if any, dominates. Theprimary difficulty lies in determining the magnitude of the reduction in stock returnvolatility that is directly attributable to corporate diversification. In essence, theregressions are comparing diversified firms, which are actually portfolios of focusedfirms, with individual stand-alone focused firms. Clearly, portfolio theory wouldpredict that diversified firms should exhibit significantly less volatility than focusedfirms. Second, there are no industry controls included. If diversified and focusedfirms tend to operate in businesses that have different levels of asymmetricinformation, then my diversification measure could simply be capturing industrydifferences. In light of the obvious difficulties of assigning a conglomerate to aparticular industry, this problem cannot be adequately addressed via industrydummies.

In an effort to address these two weaknesses, I compare diversified firms withportfolios of focused firms that operate in similar industries during the same timeperiod. More specifically, I contrast the forecast errors of diversified firms with thoseof similarly constructed portfolios of focused firms.13 This approach yields morepowerful tests of the hypotheses.

The procedure for constructing these size- and industry-matched portfolios is asfollows. For each year, I identify all the distinct industry segments of the multiple-segment firms. I then identify for each segment those focused firms in the sample thatoperate in the same industry (three-digit SIC code) as the segment during the sameforecast year. I choose as a tentative match the focused firm that is in the sameindustry as the segment and is closest in size (assets) to the conglomerate. In order toobtain a reasonable tradeoff between industry and size matching, I impose thecondition that the assets of the matching firm must be within 25% of the assets of theconglomerate. If such a match is not found at the three-digit SIC code level, I chooseamong the possible matching firms at the two-digit level. If the size criterion is notmet at the two-digit level, then the segment goes unmatched.

I have also matched the segments of conglomerates to focused firms that aresimilar to the segments in terms of both industry and size. The mean (median)difference between conglomerate forecast errors and matching-firm portfolio errorsis �0:79% (�0:66%) and this difference is significant at the 1% (1%) level. Thus, thedata appear to be inconsistent with the transparency hypothesis. However, the well-documented negative relation between firm size and analysts’ forecast errors

13 I cannot perform the matching tests for dispersion among analysts’ forecasts because these tests

require the covariance of forecasts between each of the matching firms that constitute the portfolios.


confounds this interpretation. This alternative matching algorithm does have theadvantage of explicitly recognizing that breaking up a conglomerate will result in aportfolio of generally smaller pure-play firms that will no longer benefit from anyadvantages of being part of a larger firm (e.g., more press coverage). However, theaverage matching firm under this algorithm is much smaller than the conglomerate;the mean ratio of conglomerate size to matching-firm size is 0.39. Further, anyattempt to control for size differences between conglomerates and the firmsconstituting the matching-firm portfolios is made difficult by the strong correlationbetween measures of these size differences and measures of diversification. Thus, Ireport below the results based on matching the segments of the conglomerates withfocused firms that are in similar industries as the segments but are similar in size tothe conglomerates.

The 3,814 multiple-segment firm-years in the sample include 10,620 distinctsegment-years. Suitable matches are available for 5,869 of these segment-years.However, only 1,418 multiple-segment firm-years have each of their respectiveindustry segments matched with a focused firm of appropriate size and industryaffiliation.14 These 1,418 multiple-segment firm-years include 3,698 distinct segment-years. Of these, 919 segment-years have suitable matches at the three-digit level and2,779 segment-years have suitable matches available only at the two-digit level. Themean (median) ratio of conglomerate assets to matching-firm assets is 0.99 (0.99).Thus, the focused firms are very close in size to their conglomerate counterparts. Iuse these matching firms to construct portfolios of focused firms that approximatediversified firms in terms of conglomerate size and segment industry composition.

As my matching-firm portfolio error, I calculate

PFLERR ¼XNi

i¼1

wi MFERRORi

��; ð2Þ

where wi is the ratio of segment i’s assets to the sum of segment assets, MFERRORi

is the matching-firm forecast error as a percentage of stock price, and the summationis over the N segments of the diversified firm. Thus, PFLERR is the absolute value ofthe weighted average of the signed forecast errors of matching firms, where theweights are the sizes of the segments relative to the size of the conglomerate. Bytaking weighted averages of the signed matching-firm forecast errors, the matching-firm forecast errors can offset each other if they are of opposite sign and therebyreduce the overall matching-firm portfolio forecast error. Consistent with thisinformation diversification effect, the mean absolute forecast error for the matchingfirms separately is 4.98% while the mean absolute forecast error for the portfolios

14 I lose 2,396 multiple-segment firm years because a suitable match could not be found for all segments.

I compare firms with all segments matched to firms missing matches for at least one segment and find no

significant differences in terms of forecast errors or absolute abnormal returns around earnings

announcements. However, firms with all segments matched are smaller in terms of asset size and report

fewer segments on average than firms without matches for all segments. Additionally, I run the

specifications from Tables 3 and 4 on the 1,418 conglomerate firm-years that have all segments matched.

The results using this subsample are similar to those reported in Tables 3 and 4.


formed with these matching firms (i.e., PFLERR) is 4.61%. Thus, allowing signederrors to partially offset each other results in a 0.37% reduction in average absoluteforecast errors.

The transparency hypothesis predicts that, all else equal, the forecast error for asimilarly constructed portfolio of focused firms would be smaller than that of thediversified firm. In other words, allowing errors to cancel out across focused firmssimulates the benefits of the information diversification effect that conglomeratefirms might enjoy, without introducing any of the transparency problems that mightproduce larger forecast errors for segments of conglomerates. Similarly, one can viewPFLERR as an approximation of the overall forecast error a firm could expect if itissued targeted stock for each business segment, since each segment would then haveto prepare separate GAAP conforming financial statements (e.g., see D’Souza andJacob, 2000).15 The information diversification hypothesis predicts that, all elseequal, the forecast error for a portfolio of focused firms would be very similar to thatof the diversified firm.

Table 5 reports the comparisons of diversified firms’ forecast errors and PFLERR.Due to skewness in the distributions, medians are emphasized. Consistent with theinformation diversification hypothesis, diversified firms have median forecast errorsthat are very similar in magnitude to those of their matching-firm portfolios. Recallthat by matching conglomerates’ segments with focused firms each as large as theconglomerate, I have stacked the deck in favor of the transparency effect. Thus,these findings are strongly inconsistent with the transparency hypothesis.

Surprisingly, diversified firms have mean forecast errors that are smaller inmagnitude than those of their matching-firm portfolios. However, given that I matchon size and industry, I am by design not controlling for several other firmcharacteristics known to affect forecast errors (e.g., see Table 3). Thus, I compare thedifferences between conglomerates and their matching firms in terms of R&Dexpense, leverage, intangible assets, and volatility. Differences in R&D expenditures,leverage, and intangible assets are calculated as the differences between conglom-erates and their matching-firm portfolios along these dimensions. Matching-firmportfolio R&D expenditures, leverage, and intangible assets are the weightedaverages of these variables for the matching firms, where the weights are therespective sizes of the segments relative to the conglomerates. Differences involatility are measured as the difference between the standard deviation of themarket model residuals for a conglomerate and for its matching-firm portfolio. Themarket models for both the conglomerate and its matching-firm portfolio areestimated over the 252 days preceding the conglomerate’s fiscal year-end. Matching-firm portfolio returns are the weighted average returns of the matching firms wherethe weights are the respective sizes of the segments relative to the conglomerate.

Matching-firm portfolios have significantly higher R&D expenditures and lowerleverage and intangibles. Also, conglomerates have a mean (median) abnormal

15 While it is impossible to split a conglomerate into separate firms each roughly as large as the

conglomerate, the comparison is somewhat representative of targeted stock issues where the firm continues

to operate as a conglomerate (large) firm while having multiple business lines traded separately.


return volatility of 2.21% (1.97%) while the matching-firm portfolios have a mean(median) volatility of 2.16% (1.93%); and the differences are not statisticallysignificant. Regressions (unreported) of the differences between conglomerates’forecast errors and those of their matching-firm portfolios against HERF anddifferences between conglomerates and their matching-firm portfolios in terms ofsize, volatility, R&D, leverage, and intangibles do not reveal any significant relationbetween diversification and forecast error differences after controlling for thesefactors. Thus, the reported mean difference in forecast errors in Table 5 appears tobe driven by differences between conglomerates and their matching-firm portfoliosalong dimensions not controlled for by the matching algorithm. Note that the resultsof Table 5 are robust to several changes in the matching procedures, e.g., relaxing thesize screen and/or allowing segment matches at the one-digit SIC code level.

4.4. Price impact of earnings surprises regression results

The regression results using HERF to explain absolute abnormal returns aroundearnings announcements are reported in Table 6. In the first column of Table 6, theonly explanatory variable besides year dummies is HERF. The coefficient on HERF

Table 5

Conglomerate forecast errors vs. forecast errors of similarly constructed portfolios of focused firms

1,418 multiple-segment firm-years with conglomerate size and segment industry matched focused firms for

each segment-year are included in the analysis below. ERROR is the absolute forecast error, jactual

earnings�median forecastj; scaled by the firm’s stock price five days before the earnings announcement.

PFLERR ¼ jP

wi MFERRORi j where wi is the ratio of segment i’s assets to the sum of segment assets,

MFERRORi is the value of the matching firm forecast error as a percentage of stock price, and the

summations are over the N segments of each diversified firm. Thus, PFLERR is the absolute value of a

weighted average of the signed forecast errors of matching firms where the weights are the respective sizes

of the segments relative to the conglomerate. Differences in medians are assessed using a median scores

test and the differences in means are assessed using a Wilcoxon sign-rank test.

ERROR PFLERR Differences

Year Obs. Median (%) Mean (%) Median (%) Mean (%) Medians (%) Means (%)

1985 61 2.61 5.06 3.22 7.24 � 0.61 � 2.18

1986 69 2.04 6.44 1.94 5.18 0.10 1.26

1987 84 1.60 6.73 1.56 3.74 0.04 2.99

1988 84 1.14 3.40 1.98 4.38 � 0:84nn � 0.98

1989 95 1.65 5.71 2.65 8.31 � 1.00 � 2.60

1990 159 2.06 5.81 2.21 12.63 � 0.15 � 6.82

1991 143 1.22 4.53 1.95 4.80 � 0.73 � 0:27n

1992 155 0.75 3.32 0.90 4.46 � 0.15 � 1.14

1993 182 0.93 3.31 1.00 3.75 � 0.07 � 0.44

1994 184 1.30 2.73 0.89 2.21 0.41 0.52

1995 202 1.04 2.91 1.18 3.45 � 0.14 � 0.54

Overall 1,418 1.31 4.18 1.44 4.61 � 0.13 � 0:43n



is positive and highly significant. Since increases in HERF represent increases infocus, abnormal returns around earnings announcements are significantly larger forfocused firms than for diversified firms. As in Dierkens (1991), larger reactions toearnings announcements by focused firms can indicate that the managers of focusedfirms have released substantial private information and that information asymme-tries between insiders and outsiders are larger for focused firms than for diversifiedfirms.

Given the correlation between diversification and other firm characteristicsdocumented in Table 2, additional control variables that could influence earningsannouncement abnormal returns are introduced in Columns 2 and 3. In Column 2,ERROR (defined earlier as the absolute difference between actual and forecast

Table 6

Panel regressions of earnings announcement abnormal returns on firm characteristics

The dependent variable is jARj; the absolute abnormal return for the three-day window centered on the

earnings announcement date calculated using a market model over the period 210–11 days before the

announcement. HERF is the asset-based Herfindahl Index. ERROR is the absolute forecast error, jactual

earnings�median forecastj; scaled by the firm’s stock price five days before the earnings announcement.

DISPERSION is the standard deviation of analysts’ forecasts scaled by the firm’s stock price five days

before the earnings announcement. Ln(MB) is the natural log of the ratio of the firm’s market value

(market value of equity plus the book value of total assets minus the book value of equity) to the firm’s

book value of total assets at the prior fiscal year-end. Ln(MVE) is natural log of the market value of a

firm’s equity at the end of the previous fiscal year. LEVG is the ratio of long-term debt and debt in current

liabilities to total assets. The regressions also include dummy variables for years 1986–1995. The estimated

coefficients for the year dummies are not reported below. t-Statistics based on the White-adjusted standard

errors are in parentheses ðN ¼ 12; 282Þ:

Regression (1) (2) (3)

INTERCEPT 1:80nnn 1:56nnn 1:37nnn

(10.19) (8.74) (7.40)

HERF 2:61nnn 2:42nnn 2:39nnn

(17.64) (16.57) (15.04)

ERROR 0:06nnn 0:46nnn

(7.66) (5.45)

DISPERSION 0:14nnn 0:25nnn

(3.52) (5.58)

HERF ERROR � 0:08nnn

ð�2:96ÞDISPERSION ERROR � 0:01nnn

ð�3:45ÞLn(MB) ERROR 0:09nnn

(4.67)

Ln(MVE) ERROR � 0:03nnn

ð�4:32ÞLEVG ERROR � 0:07n

ð�1:70Þ

Adjusted R2 0.03 0.05 0.06



earnings, deflated by the stock price five days before the earnings announcementdate) and DISPERSION (the standard deviation of analysts’ forecasts, deflated bythe stock price five days before the earnings announcement date) are added. Thecoefficient on ERROR (i.e., the ERC) is positive and significant consistent withbigger earnings surprises eliciting larger abnormal returns. The coefficient onDISPERSION is also positive and significant.

In Column 3 of Table 6, I add the interaction terms intended to control for thedeterminants of the ERC. The coefficient on HERF remains positive and significant,consistent with larger reactions to earnings announcements by focused firms. Also,the coefficient on the product of ERROR and HERF is negative, indicating thatgreater diversification is associated with slightly larger ERCs. One interpretation ofthese results is that it is easier for outsiders to forecast the earnings of diversifiedfirms and, when managers of diversified firms reveal actual earnings, this newinformation is capitalized in the stock price to a greater extent than the earningssurprises of focused firms. As previously documented in the ERC literature, firmswith better growth opportunities have higher ERCs while larger firms, firms withgreater leverage (risk), and firms with greater dispersion among analysts’ forecastshave lower ERCs (e.g., see Collins and Kothari, 1989; Teoh and Wong, 1993).

I assess the robustness of the reported regression results by replicating theregressions for various subsamples of the data (e.g., diversified firms, firms thatsurvive at least five years, firms above/below median size, firms that do not issuemanagement forecasts, etc.). Also, to mitigate potential problems with errors beingcross-sectionally correlated, I estimate the regression equations using the Fama andMacBeth (1973) methodology. Finally, rather than absolute abnormal returns, I usesquared abnormal returns. The primary advantage of this approach is that squaredVOLATILITY could be used as a weight in a weighted least squares regression. Theresults using these alternative approaches are similar to those reported in Table 6.16

5. Conclusion

While improved transparency is an oft-cited and potentially relevant benefit ofincreased focus, I find that, on balance, diversified firms do not exhibit higher levelsof asymmetric information than focused firms. This relation holds over time andafter controlling for other factors expected to affect asymmetric information. Theseresults suggest that while transparency is a concern for some conglomerates, it is byno means a general problem.

My results can help to shed some light on a nagging question associated with thecorporate diversification literature, namely, why so many firms remain diversified in

16 As an additional robustness check, I compare the ratio of announcement to nonannouncement

abnormal return variance as a measure of earnings informativeness following Korajczyk et al. (1991). A

comparison (Wilcoxon rank-sum test) of these ratios across firm type reveals that diversified firms have

significantly smaller increases in return variance around earnings announcements than focused firms. This

is consistent with conglomerates’ earnings being easier to forecast than those of focused firms.


light of the evidence that diversification appears to be, on average, a value-reducingstrategy. Denis et al. (1997b) appeal to problems associated with the separation ofownership and control. However, information concerns certainly could impact afirm’s choice to remain diversified rather than refocus. The findings of this papercould in part explain the reluctance of many diversified firms to pursue a morefocused organizational form.

References

Alford, A., Berger, P., 1999. A simultaneous equations analysis of forecast accuracy analyst following, and

trading volume. Journal of Accounting, Auditing, and Finance 14, 219–240.

Ali, A., Klein, A., Rosenfeld, J., 1992. Analysts’ use of information about permanent and transitory

earnings components in forecasting annual EPS. The Accounting Review 67, 183–198.

Atiase, R., 1985. Predisclosure information, firm capitalization, and security price behavior around

earnings announcements. Journal of Accounting Research 23, 21–36.

Barth, M., Kasznik, R., McNichols, M., 1998. Analyst coverage and intangible assets. Unpublished

Working Paper, Stanford University, Stanford, CA.

Beaver, W., Clarke, R., Wright, W., 1979. The association between unsystematic security returns and the

magnitude of forecast errors. Journal of Accounting Research 17, 316–340.

Bhide, A., 1990. Reversing corporate diversification. Journal of Applied Corporate Finance 5, 70–81.

Bhushan, R., 1989. Firm characteristics and analyst following. Journal of Accounting and Economics 11,

255–274.

Brennan, M.J., Subrahmanyam, A., 1995. Investment analysis and price formation in securities markets.

Journal of Financial Economics 38, 361–381.

Collins, D., Kothari, S., 1989. An analysis of intertemporal and cross-sectional determinants of earnings

response coefficients. Journal of Accounting and Economics 11, 143–181.

Comment, R., Jarrell, G., 1995. Corporate focus and stock returns. Journal of Financial Economics 37,

67–87.

Denis, D.J., Denis, D.K., Sarin, A., 1997a. Managerial incentives and corporate diversification strategies.

Journal of Applied Corporate Finance 10, 72–80.

Denis, D.J., Denis, D.K., Sarin, A., 1997b. Agency problems, equity ownership, and corporate

diversification. Journal of Finance 52, 135–160.

Diamond, D., Verrecchia, R., 1991. Disclosure, liquidity, and the cost of capital. Journal of Financial

Economics 19, 69–90.

Dierkens, N., 1991. Information asymmetry and equity issues. Journal of Financial and Quantitative

Analysis 26, 181–200.

D’Souza, J., Jacob, J., 2000. Why firms issue targeted stock. Journal of Financial Economics 56, 459–483.

Dunn, K., Nathan, S., 1998. The effect of industry diversification on consensus and individual analysts’

earnings forecasts. Unpublished Working Paper, Georgia State University, Atlanta, GA.

Easterwood, J.C., Nutt, S., 1999. Inefficiency in analysts’ earnings forecasts: systematic misreaction or

systematic optimism? Journal of Finance 54, 1777–1779.

Easton, P., Zmijewski, M., 1989. Cross-sectional variation in the stock market response to accounting

earnings announcements. Journal of Accounting and Economics 11, 117–141.

Elton, E., Gruber, M., Gultekin, M., 1984. Professional expectations: accuracy and diagnosis of errors.

Journal of Financial and Quantitative Analysis 19, 351–363.

Fama, E., MacBeth, J., 1973. Risk, return, and equilibrium—empirical tests. Journal of Political Economy

81, 607–636.

Fee, C., Thomas, S., 2001. Corporate diversification and asymmetric information: evidence from stock

market trading characteristics. Unpublished Working Paper, Michigan State University, East Lansing,

MI.


Gilson, S., Healy, P., Noe, C., Palepu, K., 2000. Corporate focus and the benefits from more specialized

analyst coverage. Unpublished Working Paper, Harvard Business School, Cambridge, MA.

Gorton, G., Pennacchi, G., 1993. Security baskets and index-linked securities. Journal of Business 26,

1–27.

Habib, M., Johnsen, D.B., Naik, N., 1997. Spinoffs and information. Journal of Financial Intermediation

6, 153–176.

Hadlock, C., Ryngaert, M., Thomas, S., 2001. Corporate structure and equity offerings: are there benefits

to diversification? Journal of Business 54, 613–635.

Haw, I., Jung, K., Ruland, W., 1994. The accuracy of financial analysts’ forecasts after mergers. Journal of

Accounting, Auditing, and Finance 9, 465–486.

Jacquemin, A., Berry, C., 1979. Entropy measure of diversification and corporate growth. The Journal of

Industrial Economics 27, 359–369.

Korajczyk, R., Lucas, D., McDonald, R., 1991. The effect of information releases on the pricing and

timing of equity issues. The Review of Financial Studies 4, 685–708.

Krishnaswami, S., Subramaniam, V., 1999. Information asymmetry, valuation, and the corporate spin-off

decision. Journal of Financial Economics 53, 73–112.

Lang, M., Lundholm, R.J., 1996. Corporate disclosure policy and analyst behavior. The Accounting

Review 71, 467–492.

Lang, L., Stulz, R., 1994. Tobin’s q, corporate diversification, and firm value. Journal of Political

Economy 102, 1248–1280.

Montgomery, C., 1994. Corporate diversification. Journal of Economic Perspectives 8, 163–178.

Nanda, V., Narayanan, M., 1999. Disentangling value: financial needs, firm scope, and divestitures.

Journal of Financial Intermediation 8, 174–204.

O’Brien, P., 1988. Analysts’ forecasts and earnings expectations. Journal of Accounting and Economics

10, 53–83.

Piotroski, J., 1999. Discretionary segment reporting decisions and the precision of investor beliefs.

Unpublished Working Paper, University of Chicago.

Subrahmanyam, A., 1991. A theory of trading stock index futures. Review of Financial Studies 1, 17–51.

Swaminathan, S., 1991. The impact of SEC mandated segment data on price variability and divergence of

beliefs. The Accounting Review 66, 23–41.

Teoh, S., Wong, T., 1993. Perceived auditor quality and the earnings response coefficient. The Accounting

Review 68, 346–366.

Zuta, S., 1998. Diversification discount and targeted stock: theory and evidence. Unpublished Working

Paper, University of Maryland, College Park, MD.


Documents

Firm diversification and asymmetric information: evidence from analysts’ forecasts and earnings announcements